Join our Team

About this opportunity

How do humans anticipate risks before hazardous events, such as while driving anticipating when an accident can happen? We usually use our previous experiences that are collection of episodic memories. Can AI systems learn from previous experiences in a similar way using episodic memory mechanisms? As intelligent systems evolve beyond perception and prediction, episodic memory is emerging as a key capability for enabling proactive reasoning and decision-making.

At Ericsson, we are exploring how memory-augmented world models can improve behavior prediction and assistance for humans, such as accident anticipation in immersive traffic simulations. This thesis investigates how different forms of memory can help AI systems better understand the environment, anticipate hazards, and provide explainable predictions. Using an existing interactive VR-based traffic simulation platform, you will explore how past experiences can be retrieved and used to support prediction of future events and risky situations.

What you will do

This Master Thesis builds upon Ericsson's ongoing research in AI, world models, digital twins, XR/VR, and human-centric intelligence. The goal is to investigate whether episodic memory mechanisms can improve behavior prediction and accident anticipation in immersive urban traffic environments.

You will have the freedom to define areas of exploration within the project. These could include, but are not limited to:

• Designing and implementing memory-augmented behavior prediction models.

• Evaluating three complementary episodic memory mechanisms:

  • Personal memory based on previous experiences of individual road users.
  • Location-based memory capturing accident patterns and risks associated with specific locations.
  • Situation-based memory representing previously observed traffic events and hazardous scenarios.

• Integrating simulation data such as trajectories, interactions, traffic context, hazardous events, and gaze information into a retrieval-based prediction framework.

• Comparing AI-based risk prediction with human attention and decision-making patterns through user studies.

• Investigating explainability and interpretability of memory-based prediction systems.

• Exploring how memory-augmented world models can support safer and more intelligent future systems.

The skills you bring

• Knowledge in Computer Science, Electrical Engineering, Artificial Intelligence, or a related field.

• Experience with Python programming.

• Interest in machine learning, deep learning, and AI research.

• Familiarity with PyTorch or TensorFlow.

• Understanding of data analysis and software engineering practices.

• Bonus: experience with VR/XR development, Unity or Unreal Engine, computer vision, simulation environments, eye-tracking data, retrieval systems, digital twins, or autonomous driving research.

• Curiosity, creativity, and willingness to explore new research ideas.

Preferred Background

This thesis is particularly suitable for students who enjoy combining machine learning, simulation, human behavior analysis, and user-centered experimentation to develop next-generation intelligent systems.