With the grace and blessings of God, the master’s thesis defense of the student Abbas Nazim Kazem was successfully conducted on Wednesday, August 6, 2025.

The thesis, entitled:

"Deep Learning-Based on Path Planning Algorithm for Mobile Robots"

was supervised by Asst. Prof. Dr. Mohammed Sabri Salim, and the student was awarded the Master’s degree with a grade of Excellent.

sincere gratitude is extended to the members of the examination committee for their valuable scientific feedback and contributions:

Prof. Dr. Anus Qusay  Hashem – Chair

Asst. Prof. Dr. Ameen Dawood Salman – Member

Asst. Prof. Dr. Ahmed Faeq – Member

We also appreciate the efforts of the scientific reviewers:

Dr. Fatima Shamsuddin Abdul Sattar

Asst. Prof. Dr. Ahmed Raouf Nasser

As well as the linguistic reviewer:

Lecturer Zeena Kamal Ibrahim

Special thanks are extended to Dr. Ali Abdulrahman, the academic coordinator of the Department of Electronic and Communications Engineering, along with the department faculty members and Ms. Raghad Fawzi for their continuous support and assistance.

The thesis introduces a new hybrid path-planning method for mobile robots called PRM-DDPG. It combines the global planning power of probabilistic roadmaps (PRM) with the local adaptability of deep reinforcement learning (DRL). PRM-DDPG combines the best parts of both sampling-based and learning-based methods, unlike other systems that only use one or the other. PRM makes waypoints quickly across big areas, and Deep Deterministic Policy Gradient (DDPG) lets the robot avoid obstacles in real time and optimize its path smoothly. This combination solves important problems with each method on its own, like PRM's poor performance in changing environments and DDPG's high cost for global planning. It creates a strong, scalable solution for autonomous navigation.

Simulation results indicate that PRM-DDPG outperforms traditional methods like PRM and RRT*, as well as machine learning options like ID3QN and optimizations methods like GA, in terms of shorter paths, quicker times, and better obstacle avoidance, especially in complex settings. For example, PRM-DDPG finds the shortest path (27.0182 m) with the fewest turns (6 corners), while ID3QN and GA find longer, less efficient paths. The algorithm is better than classical or solely learning-based techniques because its trajectories are smoother and it converges faster.

PRM-DDPG helps mobile robots get closer to better, more adaptable, and real-world usable navigation systems by connecting global and local planning.