Abstract:
The problem of state estimation of the mobile robot's trajectory being a nonlinear one, the
intent of this thesis is to go beyond the realm of the basic Extended Kalman Filter(EKF)
and study the more recent nonlinear Kalman Filters for their application to the problem at
hand. The various lters that I employ in this study are:
• Extended Kalman Filter(EKF)
• Iterated Extended Kalman Filter (IEKF)
• Unscented Kalman Filter(UKF) and its various forms and alternate editions
The Robot is given di erent trajectories to run on and the performance of the lters on
each of these trajectories is observed. The intensity of process noise and measurement noise
are also varied and the e ect they have on the estimates studied.
The study also provides a comparison of the computational costs involved in each of the
lters above. Pre-established numerically e cient and stable techniques to lower the computational
costs of the UKF are employed and their performance in accuracy and computational
time is observed and noted.
The UKF is proven to be a better lter in terms of accuracy for the non-linear cases such as
inertial navigation systems, but this thesis tests speci cally for the system dynamics of the
Mobile Robot. The results that are obtained are quite contrasting to the otherwise belief
that the UKF should give better accuracy.
