Comparative Study of Nonlinear Attitude Estimation in GNC Testbeds for Collegiate Self-Landing Rockets

When constructing a device that requires detail on its orientation in a 3D space, an inertial measurement unit (IMU) is a widely used solution. This commonly found sensor in the navigation space contains a gyroscope, accelerometer, and magnetometer, but two issues arise in IMUs when considering orientation calculation. The first is the natural noise generated from hardware electrical systems, and the second is the accumulated error when independently integrating a component into navigation equations. For instance, the double integration of noisy acceleration from the accelerometer produces an inaccurate estimate of our device's position, leading to a significant increase in error over time. The solution is to utilize filters and algorithms that incorporate every part of the nine-axis IMU to help reduce noise from their components and avoid the problem of the previously mentioned drifting. Here we present the Complementary, Fourarti, Extended Kalman Filter (EKF), Madgwick, and Mahony nonlinear estimation algorithms and compare their accuracy, when tuned, in predicting a predefined rocket path. We conclude that the Madgwick filter performed the best for our dataset, holding the smallest magnitude error. Using these results, we will proceed with the Madgwick filter in the development of our Monoprop UAV before implementing the algorithm into our rocket.