Date of Graduation


Document Type


Degree Name

Master of Science in Mechanical Engineering (MSME)

Degree Level



Mechanical Engineering


Uche Wejinya

Committee Member

Mark E Arnold

Second Committee Member

Po-Hau Huang


Deep Learning;Dynamic Modeling;Mobile Robotics;Nonlinear Systems Control;Robust Control


The problem of balancing and autonomously navigating a two-wheel mobile robot is an increasingly active area of research, due to its potential applications in last-mile delivery, pedestrian transportation, warehouse automation, parts supply, agriculture, surveillance, and monitoring. This thesis investigates the design and control of a two-wheel balancing mobile robot using three different control strategies: Proportional Integral Derivative (PID) controllers, Sliding Mode Control, and Deep Q-Learning methodology. The mobile robot is modeled using a dynamic and kinematic model, and its motion is simulated in a custom MATLAB/Simulink environment. The first part of the thesis focuses on developing a dynamic and kinematic model for the mobile robot. The robot dynamics is derived using the classical Euler-Lagrange method, where motion can be described using potential and kinetic energies of the bodies. Non-holonomic constraints are included in the model to achieve desired motion, such as non-drifting of the mobile robot. These non-holonomic constraints are included using the method of Lagrange multipliers. Navigation for the robot is developed using artificial potential field path planning to generate a map of velocity vectors that are used for the set points for linear velocity and yaw rate. The second part of the thesis focuses on developing and evaluating three different control strategies for the mobile robot: PID controllers, Hierarchical Sliding Mode Control, and Deep-Q-Learning. The performances of the different control strategies are evaluated and compared based on various metrics, such as stability, robustness to mass variations and disturbances, and tracking accuracy. The implementation and evaluation of these strategies are modeled tested in a MATLAB/SIMULINK virtual environment.

Included in

Robotics Commons