Student Theses

Student Thesis Procedure

To start your thesis with the SIR Lab, follow the given procedure linked below!

Open Thesis Topics

Perception and Analysis of Humans for Expressive Interactive Robots

Master Thesis

Contact: Till Hielscher (till.hielscher@ki.uni-stuttgart.de)

Motivation:
In the context of human-robot interaction (HRI), the perception of
humans is a central topic and a basis for dealing with non-verbal social
signals like gestures, mimicry, posture, paralinguistic cues and similar.

Task:
Working with pose and face tracking systems, create a framework
that allows a robot to segment, learn, and generalize observed sequences
of demonstrated human expressive motion. Generate open-ended
non-verbal interactive robot skills based on detected social signals

Keywords:
Human-Robot Interaction, Visual Human Analysis, Expressive Motion
Generation, Character Animation

3D Scene Graphs for Multi-Modal Visual Localization

Master Thesis

Contact: Dennis Rotondi (dennis.rotondi@ki.uni-stuttgart.de)

Motivation:
3D Scene Graphs are gaining significant attention in the robotics
community due to their hierarchically structured representation.
However, performing loop closures and localization semantically
within these graphs remains a challenging problem

Problem Description:
The goal is to identify specific portions of the graph given
visual or textual inputs by leveraging its hierarchical and
multimodal structure for precise spatial understanding and
localization

Keywords:
Multimodality, Deep Learning, 3D Scene Graphs

3D LiDAR SLAM with Implicit Mapping

Master Thesis

Contact: Fabio Scaparro (fabio.scaparro@ki.uni-stuttgart.de)

Motivation:
Implicit reconstruction is used in other sensor modalities (NERFs,
Gaussian Splats for RGB and RGB-D). Straightforward application
of those algorithms may not be suitable for online, real-time SLAM,
because of high computational costs and lack of elasticity. Also,
LiDAR applications have not been explored as much as cameras.

Problem Description:
The goal is to implement or adapt a 3D LiDAR SLAM pipeline that
works real time with implicit mapping and conduct qualitative
experiments on real data.

Keywords:
3D LiDAR SLAM, NeRFs, Neural Point Fields,
Gaussian Splatting