| Roll | Pitch | Yaw |
|---|---|---|
 |
 |
 |
This repostory estimates the orientation (roll, pitch and yaw) given IMU data using a unscented kalman filter. Ground truth was found out using a vicon mocap setup.
- Numpy
- Scipy
- Matplotlib
The main approach taken can be found in this paper. 2 approaches were taken to the state size, a state with 7 elements (4 for orientation and 3 for angular velocity) and a state with 4 elements (for orientation). The 7 state version still has to be tuned more but the 4 state has been tuned for the given datasets.
Below are some results using the first dataset. The roll and pitch are on point, while the yaw drifts. This is a limitation with estimating yaw using only an IMU.
| 7 state | 4 state | |
|---|---|---|
| Roll |  |
 |
| Pitch |  |
 |
| Yaw |  |
 |