Bachelor and Master Theses

The proposed thesis explores the vital field of synthetic data generation using cutting-edge 3D simulators (Unreal Engine 5, Unity or Blender) to address the critical challenge of testing and validating machine learning (ML) decision systems.

In today's AI-driven landscape, ML models play a pivotal role in numerous domains, from autonomous vehicles to healthcare, from autonomous machines in construction yards to quality control. Ensuring the reliability, robustness, and safety of these ML systems is paramount.

However, obtaining vast and diverse real-world data for testing purposes is expensive, time-consuming, and often insufficient to cover all potential scenarios. This limitation necessitates the exploration of synthetic data generation as a viable solution.

An alternative that is gaining ground in comparison to classic simulation fields such as Matlab or Simulink, for example, are video game engines. Gaming platforms, with their advanced rendering capabilities and physics simulations, offer an ideal platform for creating highly realistic virtual environments.

The thesis proposal aims to delve into the development of methodologies and techniques for harnessing Simulators potential to generate synthetic data that accurately represents complex real-world scenarios.

By focusing on this innovative approach, the research endeavors to advance the field of ML system testing, ensuring that AI decision systems are robust, safe, and well-prepared to handle diverse and challenging environments.

What will you do?

·         Utilize simulators to import, modify, or create models of objects or environments for simulation scenarios.

·         Leverage these simulators to generate synthetic data, which will be instrumental in testing and training ML models.

·         Employ 3D simulators to rigorously evaluate safety and security requirements in ML image decision systems.

What will you learn?

·         Basics on 3D modeling (Blender, Unreal Engine, Unity)

·         Control a 3d simulation by scripting

·         Train a ML model based on real and synthetic dataset you generate

·         Methodology in developing, testing, and validating ML decision systems models

C#, Python, or C++; background in machine learning; interested to work with game engines (like Carla, Unity ,etc.)

Saab and Volvo