On Thursday, May, 21st ,2026 the department of Mechatronics engineering at Al-Khwarizmi College of engineering held a master thesis defense titled “Humanoid robot locomotion using reinforcement learning for walking challenges”,
for the master candidate (Anas Yasin Abbas) at the thesis defense hall, where a degree of passing with distinction was awarded for his research findings.
The committee is composed of the following members:
- Professor Dr. Ali Hussein Muri, Chair
- Assistant Professor Dr. Mazin Ismail Jabbar, Examiner
- Assistant Professor Dr. Ahmed Mahrus Raghib, Examiner
- Assistant Professor Dr. Yarub Omar Naji, Supervisor
This thesis aimed to process one of more complicated cases related to humanoid robots in terms of bipedal locomotion in a stable and adaptable manner under diverse environmental conditions. Traditional approaches, such as Zero Moment Point (ZMP), faced many limitations in the case of balancing, computational cost, and adaptability, which make these systems hard to implement in real-time embedded systems.
Therefore, this thesis introduces a hybrid control architecture that combines model-based control and reflex-inspired control to enhance robot stability and efficiency during locomotion. In the simulation section, a model-based control platform optimized by combining a preview control approach with a discrete algebraic Riccati equation (DARE), linear inverted pendulum model (LIPM), and inverse kinematics (IK) to adapt foot placement. In addition, cubic spline interpolation is employed to optimize leg trajectories and make the transitions between joints smoother. This platform was implemented on a 12 DOF biped robot in the PyBullet simulator on flat terrains.
The results demonstrated that the proposed system maintains the desired ZMP inside the robot support polygon with a small ZMP-error reaching 0.03 m, high frequency control at 240 Hz, and requires only 33% of CPU usage. In addition, the system recovers external forces equal to 10 N within 0.8 s, improving its robustness and performance in real-time.
In the experimental part, a reflexive-based controller is implemented with a piezoelectric sensor for gait adaptation on slippery terrain. A fixed threshold was used with an Exponential Moving Average (EMA) filter for the classification of terrains. The robot adjusts the stance duration and velocities dynamically according to surface conditions (rough or slippery). On a rough surface, the robot takes a shorter stance duration of about 0.2 s and an angular velocity of 125°/s. On the other hand, in the case of a slip surface, the system takes a longer stance duration and decreases angular velocity (0.38 s, 53°/s), demonstrating effective slip adaptation.
The final results showed that the proposed hybrid control architecture provides a practical and cost-effective solution with high computational efficiency for achieving a stable bipedal locomotion in real time, thereby enhancing opportunities of employing humanoid robots in different applications such as services, education, and smart assistance.
