Attitude Representation and Kinematic Propagation for Low-Cost UAVs

Complementary filter-based algorithms for attitude estimation seek to minimize the short-comings associated with the low-cost MEMS sensors and low-power embedded processors that are frequently used in unmanned aerial vehicles. This paper explores the performance impact of three different ingredients for attitude estimation within the complementary filter framework: 1) kinematic propagation, 2) attitude representation, and 3) the assumptions on the behavior of the angular rate vector between samples. Forward Euler integration and matrix exponential propagation methods are compared within several attitude representations, namely the direction cosine matrix (DCM), quaternion, Euler angle, and the angle-axis schemes. Assumptions of constant, linear, and quadratic angular velocities between samples are also examined. The resulting algorithms are evaluated from the points of view of accuracy, computational load, and noise response. Tests using analytic inputs as well as real data within the framework of the SLUGS autopilot are performed to evaluate algorithm performance. The results of the experiments indicate that by combining matrix exponential propagation with a quaternion representation of attitude high accuracy can be obtained with a standard processor.