Estimating energy expenditure is essential in monitoring the intensity of physical activity and health status. Energy expenditure can be estimated based on wearable sensors such as inertial measurement unit (IMU). While a variety of methods have been developed to estimate energy expenditure during day-to-day activities, their performances have not been thoroughly evaluated under walking conditions according to various speeds and inclines. This study investigated IMU-based neural network models for energy expenditure estimation under various walking conditions and comparatively analyzed their performances in terms of sensor attachment locations and training/testing datasets. In this study, two neural network models were selected based on a previous study (Slade et al., 2019): (M1) a multilayer perceptron using sensor signals during each gait cycle, and (M2) a recurrent neural network using sensor signal sequences of a fixed window size. The results revealed the following: (i) the performance of the foot attachment model was the best among the five sensor attachment locations (0.89 W/kg for M1 and 1.14 W/kg for M2); and (ii) although the performance of M1 was superior to that of M2, M1 requires accurate gait detection for data segmentation by each stride, which hinders the usefulness of M2.

Inertial measurement unit (IMU)-based 3D joint angle estimation have a wide range of important applications, among them, in gait analysis and exoskeleton robot control. Conventionally, the joint angle was determined via the estimation of 3D orientation of each body segment using 9-axis IMUs including 3-axis magnetometers. However, a magnetometer is limited by magnetic disturbance in the vicinity of the sensor, which highly affects the accuracy of the joint angle. Accordingly, this study aims to estimate the joint angle using the 6-axis IMU signals composed of a 3-axis accelerometer and a 3-axis gyroscope without a magnetometer. This paper proposes a recurrent neural network (RNN) model, which indirectly utilizes the joint kinematic constraint and thus estimates joint angles based on 6-axis IMUs without using a magnetometer signal. The performance of the proposed model was validated for a mechanical joint and human elbow joint, under magnetically disturbed environments. Experimental results showed that the proposed RNN approach outperformed the conventional approach based on a Kalman filter (KF), i.e., RNN 3.48° vs. KF 10.01° for the mechanical joint and RNN 7.39° vs. KF 21.27° for the elbow joint.