Stress Testing for Performance Analysis of Orientation Estimation Algorithms

Technological advancements have seen the spread of micro-electromechanical systems (MEMSs) based sensors such as inertial measurement unit (IMU) systems. The latter provide object orientation information in numerous applications such as smartphones, robotics, automotive, drones, and many others. In more detail, IMUs based on MEMS technology are characterized by small size, low power, and low cost. Technically, orientation information is provided by appropriate orientation estimation algorithms fed with quantities measured by MEMS sensors, such as acceleration, angular velocity, and magnetic field. A preliminary work pointed out that such MEMS sensors suffer significant drifts when they have to endure non-standard operating profiles in terms of thermal conditions. However, when considering very low-cost devices, the influence of temperature on the operation of orientation estimation algorithms still remains an important research gap in the literature. To deal with this important aspect, this article proposes a specific test plan to analyze the performances of MEMS-based IMUs by considering a temperature stress test. Therefore, the effect of temperature was first analyzed on the individual sensors of the IMU (accelerometer, gyroscope, and magnetometer), then the effects of a very simple temperature compensation strategy were analyzed to highlight the importance of such step. Specifically, the effects of not compensating for temperature were studied considering a low-cost commercial IMU typically used in various fields of application. To better quantify such effects in practical applications, two well-known orientation estimation algorithms (i.e., complementary filter and attitude and heading reference system (AHRS) typically fed by IMUs output have been considered. The experimental results obtained using the custom devices under test (DUTs) show that temperature compensation is necessary for the best performance of the analyzed orientation estimation algorithms.