Master Projects

The projects and theses listed were developed by students of the international Master's programme in Artificial Intelligence under the supervision of CAIRO research professors.



This paper presents a low-cost eye-tracking system and a training game aimed to improve eye gaze control in children aged three to six, especially for those with conditions such as Dyspraxia and Autism Spectrum Disorders. The traditional method of manual evaluation of eye movement can lead to inaccurate results. Our system combines eye-tracking technology and a game-like approach to stimulate children to focus their gaze in different directions, track their gaze through the built-in camera of the laptop, along with head tracking, and provide analytics for the therapist. The system records and compares of coordinates the child’s eye movement to the coordinates of the moving object on the screen (character in the game). The results of the experiment show the potential of this system in providing more accurate assessments of the improvement of eye movement and speed over time.
This work has the potential for further development and application in the field of eye gaze treatment.

(By: C. Varghese, H. Al-Hujaily, S. Dhillon, V. Wahyudi, V. Tengguna / 2023)


Unpermuting an image is the process of reversing the effect of image permutation, which is theprocess of shuffling the pixels of an image in a random order .Deep learning techniques are particularlyuseful for this task as they are able to learn complex patterns and features from a given dataand image.This project aims to develop models that can unpermute an image using supervised andunsupervised learning architectures. The project is divided into two phases, the first one focuses ondeveloping a supervised learning model, where a neural network is trained using a labeled dataset ofpermuted images. The second and more ambitious phase focuses on developing an unsupervised deeplearning model, where the network is trained using only the permuted images without the existanceof their corresponding labels.

Full Paper

(By: M. Z. Shah Sarwar, N. Shiku / 2023)


Reinforcement learning (RL) has proven its good performance in solving challenging decisionmaking problems. Therefore, RL can be a promising solution for autonomous car to deal with complex driving scenarios. In this project, we explored the ability of RL in driving an indoor car autonomously. We also utilized SLAM for environment mapping and Wi-Fi technology for localization. The car was built using Waveshare JetRacer Kit, Nvidia Jetson Nano board, RPLidar A3, and Intel D455 camera.
Data collected from the Lidar is considered the agent’s state. The data is fed to the RL model to generate the best action the car should take in each state. The RL model is trained with over 200 episodes in a simulation environment. The experiment showed that the model can perform very well in simulation. In real-world environment, although the model lacks the ability to drive the car smoothly along a predefined path, the car is still able to avoid obstacles and walls.

Full Paper

(By: C. Bhairapu, C. Neeb, F. Mohamed, J. Tilly, I. Reddy Kachana, M. Saravanan, P. Nguyen Pham, Prof. Dr. A. Balzer / 2023)


Our study presents a comprehensive Question Answering (QA) system for coffee machine related questions. The system covers a wide range of topics such as maintenance, usage, and troubleshooting of coffee machines, offering quick and accurate answers to the users through its intuitive interface and natural language processing capabilities. The system provides a seamless experience for coffee machine owners and users to access the information they need, ensuring smooth operation of their machines and allowing them to enjoy their favorite beverages with ease. To evaluate the performance of these models, we fine-tuned a range of BERT-based Transformers on a manually created dataset of 653 question-answer pairs. In conclusion, our findings demonstrate the feasibility of using NLP Question Answering models to deliver technical answers about coffee machines, and highlight the importance of fine-tuning these models on task-specific data.

Full Paper

(By: M. Benkert, J. Schmidt, L. Rose, E. Ademo / 2023)

Thesis Supervision

Master's Thesis Supervision


Alexander Lorz:
"Development of a Neural Network for Targeted Modification of Images"

Ana Muñoz: "Transfer learning for crop mapping based on satellite imagery"

Muhammad Zakriya Shah Sarwar: "Step Detection"

In Progress

Priyanka Singh

Christina Varghese

Bachelor's Thesis Supervision


Michael Becker
: "Analysis of various communication styles on social media"