10.25911/5D77866DA8D0A
Khosravian, Alireza
State Estimation for Systems on Lie Groups with Nonideal Measurements
The Australian National University
2016
Lie groups
Observers
Geometric Methods
State Estimation
Predictors
Sensor Delay
Sensor Sampling
Symmetry
Invariance
Navigation
Localization
Attitude Estimation
Pose Estimation
GPS
Inertial Measurement Unit
Vector Measurements
Kinematic Systems
Flying Robots
SO(3)
SE(3)
The Australian National University
The Australian National University
2016-10-20
2016-10-20
2016
en
Thesis (PhD)
b40394608
http://hdl.handle.net/1885/109348
This thesis considers the state estimation problem for invariant systems on Lie groups with inputs in its associated Lie algebra and outputs in homogeneous spaces of the Lie group. A particular focus of this thesis is the development of state estimation methodologies for systems with nonideal measurements, especially systems with additive input measurement bias, output measurement delay, and sampled outputs. The main contribution of the thesis is to effectively employ the symmetries of the system dynamics and to benefit from the Lie group structure of the underlying state space in order to design robust state estimators that are computationally simple and are ideal for embedded applications in robotic systems. We address the input measurement bias problem by proposing a novel nonlinear observer to adaptively eliminate the input measurement bias. Despite the nonlinear and non-autonomous nature of the resulting error dynamics and the complexity of the underlying state space, the proposed observer exhibits asymptotic/exponential convergence of the state and bias estimation errors to zero. To tackle the output measurement delay problem, we propose novel dynamic predictors used in an observer-predictor arrangement. The observer provides estimates of the delayed state using the delayed output measurements and the predictor takes those estimates, compensates for the delay, and provides predictions of the current state. Separately, we propose output predictors employed in a predictor-observer arrangement to address the problem of sampled output measurements. The output predictors take the sampled measurements and provide continuous predictions of the current outputs. Feeding the predicted outputs into the observer yields estimates of the current state. Both methods rely on the invariance of the underlying system dynamics to recursively provide predictions with low computation requirements. We demonstrate applications of the theory with examples of attitude, velocity, and position estimation on SO(3) and SE(3). A key contribution of this thesis is the development of C++ libraries in an embedded implementation as well as experimental verification of the developed theory with real flight tests using model UAVs.