# Introduction

In this project, I calculate quaternion parameters from accelerometer readings and gyroscope readings. The quaternion parameters are then passed through a simple complementary filter. Upon obtaining the parameters of this rotation formalism, the orientation of the sensor platform (in this case, a quadcopter) is emulated on screen.

# Quaternion Basics

Quaternions are a set of four parameters utilized as a rotation and orientation formalism, often interpreted as a point along the surface of a four-dimensional hypersphere. These points are, in this project, referred to as $\begin{bmatrix} q_w \\ q_x \\ q_y \\ q_z \end{bmatrix}$.

## Quaternion Multiplication

Quaternion multiplication is associative and non-commutative. The product of two arbitrary quaternions, $q_0 = \begin{bmatrix} q_{0w} \\ q_{0x} \\ q_{0y} \\ q_{0z} \end{bmatrix}$ and $q_1 = \begin{bmatrix} q_{1w} \\ q_{1x} \\ q_{1y} \\ q_{1z} \end{bmatrix}$, are defined as \begin{multline} q_0 \bigotimes q_1 = q_{0w} q_{1w} - q_{0x} q_{1x} - q_{0y} q_{1y} - q_{0z} q_{1z} + (q_{0w} q_{1x} + q_{0x} q_{1w} + q_{0y} q_{1z} - q_{0z} q_{1y})i + (q_{0w} q_{1y} - q_{0x} q_{1z} + q_{0y} q_{1w} + q_{0z} q_{1x})j + (q_{0w} q_{1z} + q_{0x} q_{1y} - q_{0y} q_{1x} + q_{0z} q_{1w})k = \begin{bmatrix} q_{0w} q_{1w} - q_{0x} q_{1x} - q_{0y} q_{1y} - q_{0z} q_{1z} \\ q_{0w} q_{1x} + q_{0x} q_{1w} + q_{0y} q_{1z} - q_{0z} q_{1y} \\ q_{0w} q_{1y} - q_{0x} q_{1z} + q_{0y} q_{1w} + q_{0z} q_{1x} \\ q_{0w} q_{1z} + q_{0x} q_{1y} - q_{0y} q_{1x} + q_{0z} q_{1w} \end{bmatrix} \end{multline} The above result can be achieved by multiplying each elements of $q_0$ with all elements of $q_1$, keeping in mind that $i \cdot j = k$, $j \cdot k = i$, $k \cdot i = j$. Similarly, $k \cdot j = -i$, $j \cdot i = -k$, and $i \cdot k = -j$. Finally, any of the imaginary units ($i$, $j$, $k$), when multiplied by itself, is $-1$.

## Quaternion Rotation and its Rotation Matrix

Rotation of a vector, $v = \begin{bmatrix} x \\ y \\ z \end{bmatrix}$ by a quaternion, $q = \begin{bmatrix} q_w \\ q_x \\ q_y \\ q_z \end{bmatrix}$, is accomplished through prepending a $0$ to the vector to be rotated, and multiplying by the quaternion, as well as it's conjugate, $\bar q = \begin{bmatrix} q_w \\ -q_x \\ -q_y \\ -q_z \end{bmatrix}$ in the following order. \begin{equation} q'= \begin{bmatrix} q_w \\ q_x \\ q_y \\ q_z \end{bmatrix} \bigotimes \begin{bmatrix} 0 \\ x \\ y \\ z \end{bmatrix} \bigotimes \begin{bmatrix} q_w \\ -q_x \\ -q_y \\ -q_z \end{bmatrix} \end{equation}. It is worth noting that, for the above to be a valid rotation, the quaternion must have unit norm. When the above operation is performed, the equivalent rotation matrix is found to be \begin{equation} \begin{bmatrix} 1-2(q_y^2+q_z^2) & 2(q_x q_y - q_z q_w) & 2(q_y q_w + q_x q_z) \\ 2(q_z q_w + q_x q_y) & 1-2(q_x^2+q_z^2) & 2(q_z q_y - q_x q_w) \\ 2(q_z q_x - q_y q_w) & 2(q_x q_w + q_z q_y) & 1-2(q_x^2+q_y^2) \end{bmatrix} \end{equation}

## Quaternion parameters to Angular Velocity

Quaternion parameters, their time-derivatives, and angular velocoity share the following relation. \begin{align} \begin{split} w_x = 2 (q_w \dot q_x + q_z \dot q_y - q_y \dot q_z - q_x \dot q_w) \\ w_y = 2 (q_x \dot q_z + q_w \dot q_y - q_z \dot q_x - q_y \dot q_w) \\ w_z = 2 (q_w \dot q_z + q_y \dot q_x - q_x \dot q_y - q_z \dot q_w) \end{split} \end{align} From the above, and the quaternion unit-norm constraint, the following relation is derived. \begin{align} \begin{split} \dot q_w = \frac{-q_x w_x - q_y w_y - q_z w_z}{2} \\ \dot q_x = \frac{q_w w_x - q_z w_y + q_y w_z}{2} \\ \dot q_y = \frac{q_w w_y - q_x w_z + q_z w_x}{2} \\ \dot q_z = \frac{q_w w_z + q_x w_y - q_y w_x}{2} \end{split} \end{align}

# Quaternion Parameters from Accelerometer

The accelerometer returns the gravity vector in body-fixed coordinates. Thus, to find the attitude quaternion parameters of the accelerometer, the rotation quaternion between the two vectors (gravity and fixed $\begin{bmatrix} 0\\ 0\\ 1\end{bmatrix}$) should be found. The 3-D vector component of a quaternion, $\begin{bmatrix} x \\ y \\ z \end{bmatrix}$ representing such a rotation between two vectors, $v_1$ and $v_2$, can be found through the cross product of the two vectors, while the $q_w$ component can be found through the following expression: \begin{equation} q_w = \sqrt{(v_1 \cdot v_1)(v_2 \cdot v_2) + v_1 \cdot v_2} \end{equation}

# Quaternion Parameters from Gyroscope

A gyroscope outputs the angular velocity in the quadcopter's body-fixed frame. Above, we found the time-derivatives of each of the quaternion parameters. \begin{align} \begin{split} \dot q_w = \frac{-q_x w_x - q_y w_y - q_z w_z}{2} \\ \dot q_x = \frac{q_w w_x - q_z w_y + q_y w_z}{2} \\ \dot q_y = \frac{q_w w_y - q_x w_z + q_z w_x}{2} \\ \dot q_z = \frac{q_w w_z + q_x w_y - q_y w_x}{2} \end{split} \end{align} By integrating the time derivatives of the quaternion parameters, an approximation of the current quadcopter attitude quaternion can be found.

# The Complementary Filter

Because accelerometer-based quaternion parameters are more susceptible to high-frequency noise while gyroscope-based quaternion parameters are more susceptible to low-frequency drift, a complementary filter was utilized to mitigate accelerometer noise, as well as gyroscope drift. \begin{equation} q = (1-\alpha) q_{gyro} + \alpha q_{accel} \end{equation} In this particular case, an $\alpha$ value of 0.02 was used. Here's the resulting product.

One final note: If you notice, I never give the quadcopter any yaw. And that's because the accelerometer-based attitude cannot take into account any yaw. For example, relying solely on how much your head is tilted, you wouldn't be able to tell me whether your head is tilted north, or if it's tilted south. Though this shouldn't really take away from a quadcopter's ability to direct itself upwards and away from the ground, do we really want to let loose a drone with no sense of navigation?