Development and balancing control of a unicycle robot | |
| Author | Jadoon, Nabeel Ahmad Khan |
| Call Number | AIT Thesis no.ISE-24-12 |
| Subject(s) | Robotics and Automation Robotics--Design and construction |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Mechatronics and Machine Intelligence |
| Publisher | Asian Institute of Technology |
| Abstract | n 520 This thesis introduces the design, modeling, and control strategies employed in the development of a self-balancing unicycle robot. It investigates both classical (model based) and reinforcement learning (model-free) methodologies. A compact reaction wheel-based unicycle is carefully designed using Solidworks and produced using 3D printing and the integration of electronics for assembly for practical experimentation. the design of the proposed unicycle offers two unstable degrees of freedom characterized by pitch and roll, providing ground for nonlinear model control research in robotics. for the simulation study, a continuous control scheme is proposed, featuring two distinct algorithms: Linear Quadratic Regulator (LQR) in the classical domain and Deep Deterministic Policy Gradient (DDPG) in the deep reinforcement learning domain, tailored for balancing and maneuvering tasks. MATLAB is employed for classical control simulations, while Pybullet Physics Engine with Python interface is utilized for DDPG-based reinforcement learning simulations, effectively demonstrating the efficacy of the proposed control strategies. for discrete control, a proof-of-concept model based on a 2D inverted pendulum is proposed to explore self-erecting dynamics for unicycles. The performance of the control algorithms is rigorously assessed through comprehensive testing procedures, focusing on metrics such as settling time, overshoot, and robustness to external stimuli. Analysis of the results indicates that both methods demonstrate potential for balancing unicycles. However, LQR outperforms DDPG across various scenarios, showcasing greater robustness and stability, particularly concerning steady-state performance. Conversely, DDPG shows promising results and exploratory behavior yet the effective policy transfer to hardware is left for future reference. For classification, for systems with non-linear study the DRL method is suggested while the classical control methods are recommended for systems with a known dynamic, owing to their simplicity and robustness, contingent upon applicability. |
| Year | 2024 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Industrial Systems Engineering (DISE) |
| Academic Program/FoS | Mechatronics and Machine Intelligence (MMI) |
| Chairperson(s) | Manukid Parnichkun; |
| Examination Committee(s) | Mongkol Ekpanyapong;Yamakita, Masaki; |
| Scholarship Donor(s) | His Majesty the King's Scholarships (Thailand); |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2024 |