University of Twente Student Theses
Reinforcement learning control of unicycle robot
Farmehinifarahani, S. (2024) Reinforcement learning control of unicycle robot.
|
PDF
2MB |
| Abstract: | Unicycle robots, distinguished by their single-wheel design, offer a significant advantage in manoeuvrability due to their compact footprint. Maintaining stability for these robots presents a unique challenge. While the driving wheel directly controls longitudinal stabibility, achieving lateral stability requires a more intricate approach. The Moment Exchange Unicycle Robot (MEUR) addresses this challenge with an innovative design. It utilises a motor-powered reaction wheel to generate a counteracting force, ensuring lateral stability. This thesis delves into the application of reinforcement learning for achieving self-balancing control in the MEUR. Unlike traditional methods that rely on pre-defined models or hand-crafted rules, RL agents learn autonomously through repeated interactions with their environment. This "trial-and-error" learning process allows them to tackle complex and unpredictable control challenges. The ability to learn from experience and adapt to changing environments makes RL a compelling choice for various robotics applications, and the MEUR serves as a compelling case study in this regard. Through this research, the effectiveness of two prominent algorithms, namely Deep Q-learning (DQN) and Advantage Actor-Critic (A2C), is showcased in attaining stability across both roll and pitch angles of the MEUR. Building upon this foundation, the research further investigates the ability of these algorithms to control the MEUR’s movement in the longitudinal direction (forward and backward) while maintaining stability. Finally, the ability of these RL-based controllers to deal with noise and uncertainties is discussed. |
| Item Type: | Essay (Master) |
| Faculty: | ET: Engineering Technology |
| Programme: | Systems and Control MSc (60359) |
| Link to this item: | https://purl.utwente.nl/essays/98684 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page