- Invariant Kalman Filter = KF with lie algebra
- Ground reaction force => approximation of checking Coulomb cone
- Bayes filter for contact estimation requires a lot of parameter tuning
- parameter tuning can be reduced by data driven models
- underactuated dynamics = Newton Euler equations = the variation of the system momentum is equal to the external wrench applied to the robot
- wrench = torsion (wrench measurements used to estimate CoM dynamics)
- simplify models for dynamics: LIPM (linear inverse pendulum model)
complementary filter (CF):
- used when there is several sensors used for the same variable but at different frequencies
- we apply high pass filter or low pass filter regarding the sensor then fuse the output
Environment awareness:
- 2.5D maps = 2D maps with height information
Factor graph:
- BA is a special case of factor graph optimization
- BA scales linearly with the number of features whereas filtering scales cubicly
- BA relinearize at each solver step ie linearization gets better whereas filters linearize only once