Using Recurrent Neural Networks to approximate orientation with Accelerometers and Magnetometers

Many smart devices possess sensors that enable them to collect data regarding their environment and their users’ actions. For example, gyroscopes allow smart devices to determine the orientation. Functions such as exercise detection can then exploit this orientation data. However, gyroscopes are expensive and power-hungry. In contrast, accelerometers and magnetometers are cheap and relatively energy efficient. Hence, if we could use data accelerometers and magnetometers to approximate orientation through computations on the device, we can maintain the desired functionality at a lower cost. In this paper, we benchmark several Recurrent Neural Network (RNN) architectures that use accelerometer and magnetometer data to approximate orientation with reasonable accuracy.