Real-time attitude estimation based on Gradient Descent algorithm

Estimating orientation of a rigid body plays a critical role in a large part of applications like motion research, virtual reality, training and simulation. This works describes the development and evaluation of a sensor fusion filter for attitude estimation based on Gradient Descent algorithm, implemented inside an embedded system. The filter runs at 100[Hz] sampling frequency on 32 [bits] ALU, employing Quaternion algebra. The digital system waranties syncronized data acquisition and transmission. For doing this, first, tri-axial angular rate, acceleration, and magnetic field are measured in strapdown system. Then, the sensor fusion algorithm obtains an estimate of body orientation based on kinamatic model and solving an optimization problem, minimizing the error between expected and the measured fields. The system shows static accuracy lower than 0.7[degree] in 3 axis.