Embedding Artificial Intelligence onto low-power devices is a challenging task that has been partly overcome with recent advances in machine learning and hardware design. Presently, deep neural networks can be deployed on embedded targets to perform different tasks such as speech recognition, object detection or Human Activity Recognition. However, there is still room for optimization of deep neural networks onto embedded devices. These optimizations mainly address power consumption, memory and real-time constraints, but also an easier deployment at the edge. Moreover, there is still a need for a better understanding of what can be achieved for different use cases. This work focuses on quantization and deployment of deep neural networks onto low-power 32-bit microcontrollers. The quantization methods, relevant in the context of an embedded execution onto a microcontroller, are first outlined. Then, a new framework for end-to-end deep neural networks training, quantization and deployment is presented. This framework, called MicroAI, is designed as an alternative to existing inference engines (TensorFlow Lite for Microcontrollers and STM32Cube.AI). Our framework can indeed be easily adjusted and/or extended for specific use cases. Execution using single precision 32-bit floating-point as well as fixed-point on 8- and 16 bits integers are supported. The proposed quantization method is evaluated with three different datasets (UCI-HAR, Spoken MNIST and GTSRB). Finally, a comparison study between MicroAI and both existing embedded inference engines is provided in terms of memory and power efficiency. On-device evaluation is done using ARM Cortex-M4F-based microcontrollers (Ambiq Apollo3 and STM32L452RE).

Wireless sensor networks (WSNs) with energy harvesting capabilities have drawn increasing attention in the last few years, as they enable long-term monitoring applications. However, the level of power harvested is usually limited to few mW. To improve the energy efficiency of WSNs, many power management techniques have been proposed to adjust the quality of service according to the harvested energy fluctuations. As wireless communications consume a major fraction of the available energy, numerous medium access control (MAC) protocols have been proposed to minimize energy consumption, latency, and data collisions. In this paper, we present an innovative MAC protocol for energy-harvesting based WSNs exploiting ultralow-power wake-up radios. To overcome the limited range typical of wake-up radios, a multi-hop wake-up scheme based on a dual radio system is proposed enabling asynchronous communications between a base station and any node of the network while maintaining a low latency and a high energy efficiency. To reduce energy consumption, wake-up calls and data packets are transmitted using two distinct data rates. Combined with destination address decoding, using a higher data rate for data transmission also minimizes the risk of collisions. Our approach has been applied to monitoring applications composed of autonomous sensor nodes powered by indoor light energy. OMNeT++ simulation results demonstrate the benefits of our wake-up radio-based approach in terms of energy, latency, and collisions when compared with the state-of-the-art duty-cycled MAC protocols. Experiments performed with real WSN platforms equipped with a wake-up radio prototype confirm the efficiency of our approach.

Driver drowsiness is one of the major causes of accidents and fatal road crashes, causing a high human and economic cost. Recently, automatic drowsiness detection has begun to be recognized as a promising solution, receiving growing attention from industry and academics. In this letter, we propose to embed a convolutional neural network (CNN)-based solution in smart connected glasses to detect eye blinks and use them to estimate the driver’s drowsiness level. This innovative solution is compared with a more traditional method, based on a detection threshold mechanism. The performance, battery lifetime and memory footprint of both solutions are assessed for embedded implementation in connected glasses. The results demonstrate that CNN outperforms the accuracy obtained by the threshold-based algorithm by more than 7%. Moreover, increased overheads in terms of memory and battery lifetime are acceptable, thus making CNN a viable solution for drowsiness detection in wearable devices.

Human activity recognition can help in elderly care by monitoring the physical activities of a subject and identifying a degradation in physical abilities. Vision-based approaches require setting up cameras in the environment, while most body-worn sensor approaches can be a burden on the elderly due to the need of wearing additional devices. Another solution consists in using smart glasses, a much less intrusive device that also leverages the fact that the elderly often already wear glasses. In this article, we propose UCA-EHAR, a novel dataset for human activity recognition using smart glasses. UCA-EHAR addresses the lack of usable data from smart glasses for human activity recognition purpose. The data are collected from a gyroscope, an accelerometer and a barometer embedded onto smart glasses with 20 subjects performing 8 different activities (STANDING, SITTING, WALKING, LYING, WALKING_DOWNSTAIRS, WALKING_UPSTAIRS, RUNNING, and DRINKING). Results of the classification task are provided using a residual neural network. Additionally, the neural network is quantized and deployed on the smart glasses using the open-source MicroAI framework in order to provide a live human activity recognition application based on our dataset. Power consumption is also analysed when performing live inference on the smart glasses’ microcontroller.

Predictive maintenance in the car industry is an active field of research for machine learning and anomaly detection. The capability of cars to produce time series data from sensors is growing as the car industry is heading towards more connected and electric vehicles. Unsupervised anomaly detectors are therefore very adapted to process those complex multidimensional time series and highlight abnormal behaviors. We propose to use recurrent and convolutional neural networks based on unsupervised anomaly detectors with simple architectures on real, multidimensional time series generated by the car sensors and extracted from the Controller Area Network bus (CAN). Our method is then evaluated through known specific anomalies. As the computational costs of Machine Learning algorithms are a rising issue regarding embedded scenarios such as car anomaly detection, we also focus on creating anomaly detectors that are as small as possible. Using a state-of-the-art methodology incorporating a time series predictor and a prediction-error-based anomaly detector, we show that we can obtain roughly the same anomaly detection performance with smaller predictors, reducing parameters and calculations by up to $23\%$ and $60\%$, respectively. Finally, we introduce a method to correlate variables with specific anomalies by using anomaly detector results and labels.

System lifetime is the crucial problem of Wireless Sensor Networks (WSNs), and exploiting environmental energy provides a potential solution for this problem. When considering self-powered systems, the Power Manager (PM) plays an important role in energy harvesting WSNs. Instead of minimizing the consumption energy as in the case of battery powered systems, it makes the harvesting node converge to Energy Neutral Operation (ENO) to achieve a theoretically infinite lifetime and maximize the system performance. In this paper, a low complexity PM with a Proportional Integral Derivative (PID) controller is introduced. This PM monitors the buffered energy in the storage device and performs adaptation by changing the wake-up period of the wireless node. This shows the interest of our approach since the impractical monitoring harvested energy as well as consumed energy is not required as it is the case in other previously proposed techniques. Experimental results are performed on a real WSN platform with two solar cells in an indoor environment. The PID controller provides a practical strategy for long-term operations of the node in various environmental conditions.