Skip to content

Latest commit

 

History

History
29 lines (25 loc) · 970 Bytes

invariant_KF_VISLAM.md

File metadata and controls

29 lines (25 loc) · 970 Bytes

Invariant Kalman Filtering for Visual Inertial SLAM

Author: Martin Brossard

Year: 2018

Notes:

  • Applied recently discovered UKF on Lie Group (UKF-LG)
  • Composite manifold $\chi \in SE_{2+p}(3)$ with pose, velocity and lmks: $$ \chi=\left[\begin{array}{cccccc} \mathbf{R} & \mathbf{v} & \mathbf{x} & \mathbf{p}1 & \cdots & \mathbf{p}p \ \mathbf{0}{2+p \times 3} & & \mathbf{I}{2+p \times 2+p} \end{array}\right] . $$
  • UKF spares the user from jacobians computation
  • compare left and right multiplication: right is better
  • $\dot{R} = R (\omega -b_{\omega})_{\times}$
  • Random variable on LG with left multiplication by the mean: $$ \chi=\bar{\chi} \exp (\boldsymbol{\xi}), \boldsymbol{\xi} \sim \mathcal{N}(\mathbf{0}, \mathbf{P}) $$
  • Can then define a left-UKF-LG and a right-UKF-LG
  • Generates Sigma Points in the lie algebra and then map them into the group
  • Propagation and update step
  • Simulation + 5 trajectory of EUROC
  • runtime pourri (matlab)