Design and Evaluation of Novel Attitude Estimation System Using MEMS Sensors for Indoor UAS

Abstract

Most small unmanned aerial systems in use today, employ extended Kalman filter sensor fusion algorithms in order to provide accurate estimations of attitude or orientation. These complex algorithms use measurements from GPS receivers and magnetometer sensors that can be rendered useless in GPS denied environments or areas of significant magnetic interference, such as inside buildings or other structures. The complexity of these algorithms makes them inaccessible for some researchers and hobbyists who wish to code their own attitude estimation algorithms. This complexity is also computationally expensive and requires processors that are powerful enough to operate the algorithms along with any command and control functions required by the application. In contrast, there are simple sensor fusion algorithms such as the complementary filter or linear Kalman filter, that are commonly used by hobbyists because they are relatively easy to implement and computationally lightweight. However, these methods are not as accurate as the extended Kalman filter and therefore, are not adequate for some of the emerging precision applications in aerial robotics. The goal of this research is to investigate an attitude estimation algorithm that uses two separate inertial measurement units (IMUs), each consisting of tri-axis accelerometers and tri-axis gyroscopes. This dual or twin IMU (TIMU) algorithm is compared to several common algorithms that only use one IMU, such as the complementary filter and linear Kalman filter. Analysis of a one degree of freedom experiment shows that the TIMU algorithm provides a more accurate attitude estimate. The analysis also shows that distance between the IMU and the rotating body’s center of gravity can have an inverse effect on attitude accuracy. The ability of the algorithms to provide an accurate estimate of the rate of attitude change is used as a performance metric, in addition to the accuracy of attitude estimates. The complexity of the twin IMU algorithm is kept to a minimum. It is presented in a way that can be easily programed by the layman and has a small computational footprint.