Ahmad Fahmi (2026) Optimization-based observer in enhanced Kalman filtering for feedback control of a self-balancing robot. Doctoral thesis, Universiti Teknikal Malaysia Melaka.
![[img]](http://eprints.utem.edu.my/style/images/fileicons/text.png) |
Text (24 Pages)
Optimization-based observer in enhanced_24Pages.pdf Download (538kB) |
|
Text (Full Text)
Optimization-based observer in enhanced_.pdf Restricted to Registered users only Download (6MB) |
Abstract
Accurate and robust state estimation is essential for achieving stable and responsive feedback control in self-balancing robotic systems, which are inherently nonlinear, unstable, and highly sensitive to sensor noise and external disturbances. Conventional Kalman filters, though widely used, are strongly dependent on precise parameter tuning and often degrade in performance when system dynamics or noise statistics vary. Previous optimization-based enhancements, such as the Particle Swarm Optimization (PSO)-tuned Kalman filter, have improved convergence but still suffer from premature stagnation, sensitivity to local optima, and weak constraint-handling capabilities, leading to suboptimal estimation accuracy under time-varying or noisy conditions. To address these limitations, this dissertation introduces a Fish Swarm Optimization-Based Observer (FSOB), a novel observer framework that embeds a Penalty-Based Fish Swarm Optimization Algorithm (FSOA-P) within the Kalman filtering structure. The proposed FSOA-P incorporates adaptive penalty functions to dynamically regulate constraint violations and enhance global search capability, thereby improving convergence efficiency, estimation precision, and robustness against parameter drift. Benchmark tests on standard optimization functions show that FSOA-P surpasses both conventional FSOA and PSO in terms of convergence speed, mean squared error reduction, and robustness against<br /> premature convergence. The integration of FSOA-P into the observer framework known as FSOB was validated through MATLAB simulations and real-time experiments on a twowheeled self-balancing robot. Comparative results revealed that FSOB achieved superior performance, including a reduced Root Mean Square Error (RMSE) of 0.94°, faster disturbance recovery (1.6 s on average), minimal estimation drift (1.3°/min), and a 95% disturbance recovery success rate, outperforming both the conventional Kalman filter and PSO-optimized Kalman filters. Overall, FSOB demonstrates significant improvements in adaptability, robustness, and estimation accuracy for real-time balancing control. This research contributes firstly, a theoretical advancement in penalty-based swarm optimization, secondly, a methodological innovation in observer design for nonlinear systems, and thirdly, experimental validation on embedded robotic platforms.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | Kalman filters, Particle swarm optimization, State estimation, Self-balancing robot, Fish swarm optimization |
| Subjects: | T Technology T Technology > TJ Mechanical engineering and machinery |
| Divisions: | Faculty Of Electrical Technology And Engineering |
| Depositing User: | Wizana Abd Jalil |
| Date Deposited: | 20 Jul 2026 01:34 |
| Last Modified: | 20 Jul 2026 01:34 |
| URI: | http://eprints.utem.edu.my/id/eprint/30340 |
| Statistic Details: | View Download Statistic |
Actions (login required)
 |
View Item |