Master Thesis: Multi-modal head direction estimation mechanism for neuromorphic applications

Lab: EIS lab 

Supervisor: Prof. Dr. Melika Payvand, Dr. Laura Kriener 

Collaborators: Dr. Alpha Renner and Sven Krause (Forschungszentrum Jülich) 

 

Project description: 

At the EIS lab, we are interested in solving the problems intelligent agents might face in the world using solutions inspired by our latest understanding of the brain. One such problem is autonomous navigation, where an agent (such as a drone or robot) needs to perceive its environment and its own position within it while moving and potentially interacting with the environment. One important capability required for this task is to track and integrate one's own movement relative to the environment in order to maintain a good estimate of the current position and heading direction. 

Recently, collaborators of ours developed a comprehensive model of the heading direction circuitry in the brain based on neurological recordings [1]. They hypothesize that in biological systems an elaborate combination of multiple subsystems uses both visual cues as well as angular-head velocity to reliably estimate the head direction. We now want to study whether it is feasible to mimic this biological model in our neuromorphic systems. 

 

In this thesis you will: 

• adapt and refine the high-level biological model to make it compatible with our hardware systems (e.g. port it to spiking neural networks) 

• study its behavior and properties in ideal simulations (e.g. what are the trade-offs between network sizes and resulting precision of the head direction estimate?) 

• study the impact of constraints of neuromorphic hardware (e.g. limited parameter ranges and precision, noise) and perform a feasibility study of future deployment of this model on the new generation of our MOSAIC chip [2]. 

 

What should you bring to the table: 

• Curiosity about the functional mechanism of biological systems. 

• At the same time, interest in adapting brain-like mechanisms to the constraints encountered in artificial intelligent systems. 

• Good programming skills in Python. 

• Experience with modeling spiking neural networks is a big plus. 

 

What will you get out of this: 

• Be mentored by experts in NeuroAI and neuromorphic algorithms 

• Solve challenging and significant problems with (hopefully) future applications in energy-efficient robotics 

• Gain hardware-algorithm co-design mentality by interacting with both algorithm and circuit designers on a regular basis 

 

[1] Krausse, Sven, et al. "Parallax error in the head-direction system indicates simple cue-anchoring mechanism." BioRxiv (2025): 2025-04.  

[2] Dalgaty, Thomas, et al. "Mosaic: in-memory computing and routing for small-world spike-based neuromorphic systems." Nature Communications 15.1 (2024): 142.