Sensory information processing in robot skins have till date relied on a centralized approach with reception and signal transduction separated from the location of decision-making. These approaches have been found wanting with respect to constantly shuttling large amounts of data from the periphery to the center, resulting in wiring, latency, fault tolerance and robustness issues. We thus envision a decentralized approach with intelligence shifted to the location of the sensing nodes as a novel neuromorphic approach to tackle the issues of information processing and learning in robotic skins.
In this work, we configure memristive devices as peripheral processors to enable energy-efficient edge computing for sensory signal computing. In comparison to the very recent implementations of artificial afferent nerves with physically separate signal transduction and processing, we propose a novel decentralized scheme and demonstrate decision making at the sensor node as a viable solution to address the peripheral sensory signal processing in robotics.
We develop a proof-of-concept system comprising of artificial nociceptors that respond to pain, coupled with memristive learning synapses and neurons to associate pressure and pain perception (Figure 1). Nociception is implemented by Satellite Threshold Adjusting Receptors (STARs). The associative learning is implemented in satellite learning modules (SLMs) near the sensing nodes, composed of satellite weight adjusting resistive memories (SWARMs) and satellite spiking neurons (SSNs). We demonstrate a three-tier decision making process flow- nociceptors identify and filter noxious information based on short-term temporal correlations, synapses associatively learn patterns in sensory signals with noxious information and neurons integrate synaptic weights.
Figure 1. Analogy between biological and our prototypical artificial nervous system
Most importantly, although sensorized artificial synapses approaching human skin-like performance in terms of mechanical sensing and form factor have been very recently demonstrated, the ability to repeatably self-heal neuromorphic circuit elements have not been demonstrated yet. We demonstrate complete functional and mechanical self-healing of our devices upon injury and the associative learning within the learning modules enables good signal integrity even if the nociceptor is damaged after learning, enhancing the fault tolerance (Figure 2). The self-healing capability of our intelligent devices open up the possibility that robots may one day have an artificial nervous system that can repair itself. This ability is hitherto not demonstrated for hardware neuromorphic circuits and is timely especially with the future of electronics and robotics going soft.
Figure 2. Self-healable neuromorphic elements. a Schematic of the mechanism of healing. b SEM images of the self-healing ion gel dielectric. c healing behaviour of our artificial nociceptors and d artificial synapses.
Please check out our recent work published on Nature Communications: “Self healable neuromorphic memtransistor elements for decentralized sensory signal processing in robotics” at the link: https://doi.org/10.1038/s41467-020-17870-6.