LuoXubo / IMU-Aided-Odometry

Odometry Trajectory Recursion Based on IMU Integral Interpolation.

IMU-Aided-Odometry

Odometry Trajectory Recursion Based on IMU Integral Interpolation.

Notice

• This repository is heavily based on ETH's odom predictor package https://github.com/ethz-asl/odom_predictor and Abin1258's https://github.com/Abin1258/imu_to_odom.
• Compared with the original versions, we have increased the monitoring of the odometry topic to eliminate the cumulative error.
• We fixed bugs in Abin1258's repo in the transformation from the orientation matrix to the quaternion.

Overall I/O

Input

• High frequency IMU topic (50Hz)
• Low frequency Odometry topic (1Hz)

Output

• High frequency pose trajectory after IMU estimation and Odometry correction

Method

Orientation estimation

$$ \Delta R ={ {\begin{bmatrix} 0& -w_z\times\Delta t& w_y\times \Delta t\

w_z\times\Delta t& 0& -w_x\times\Delta t\

-w_y\times\Delta t& w_x\times\Delta t& 0 \end{bmatrix}} }, $$

$$ \sigma = \sqrt{w_x^2 + w_y^2 + w_z^2} \times \Delta t, $$

$$ R_t =R_{t-1}\times (\mathbf{I}_3 + \frac{sin(\sigma)}{\sigma}\times \Delta R - \frac{1-cos(\sigma)}{\sigma^2}\times \Delta R^2), $$

$$ v_t = v_{t-1} + \Delta t \times (R_t \times a_{imu} - g), $$

$$ pos_t = pos_{t-1} + \Delta t \times v $$

Drift correction

• Use the pose information from Odometry topic to cover the current pose vector.
• Velocity correction:
  • Assume that $|v_{mid}| = |v_{t}|$
  • $v_{mid} = \frac{Pos_{t} - Pos_{t-1}}{\Delta t}$
  • $v_t = R_t\times R_{t-1}^{-1}\times v_{mid}$

Experimental verification

• RMSE/Length = 0.10156%