Self-powered edge intelligence using nanogenerators for wearable IoT

Edge computing is transforming the world's IoT systems. Sensing, reasoning, and decision-making are directly assigned to data-gathering devices at the edge of the network. In this trend, flexible nanogenerators (NGs) have become a prominent research focus, especially piezoelectric (PENG) and triboelectric (TENG) devices, which provide energy collection and self-powering capabilities for wearable devices and IoT nodes. However, the characteristics of flexible NGs also limit their practical application. Some important difficulties include low output power, intermittent power, and environmental sensitivity, which complicate deployment. One of the candidates expected to promote a breakthrough in NG technology is artificial intelligence. Machine learning (ML) and deep learning (DL) algorithms are becoming practical tools to support the solutions to problems encountered. This review examines the research on flexible NG for wearable systems in recent years. NGs are quickly evaluated based on their mechanism, fabrication materials, and placement within a complete wearable system. Then, we focus on analyzing the combination of NG and embedded AI on small, resource-constrained wearable devices. Traditional and modern AI algorithms, along with key optimization techniques, are employed to deploy ML and DL models based on signals collected from NG, such as quantization, pruning, feature selection, on-device preprocessing, or low-sampling rate optimization. More specifically, some platforms that utilize embedded AI are reviewed and evaluated, with guidance on the trade-off between performance and complexity. The platforms are especially suitable for researchers who wish to conduct multidisciplinary research combining NG with embedded AI systems, regardless of their AI background. Finally, we identify the key challenges today and outline a roadmap towards robust, self-sustaining, privacy-preserving edge NG-AI systems for wearables.