Christoph Würsch

Prof. Dr. Christoph Würsch

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 June 2021 in Education Sciences
Verified
Journal Article
Open access
PDF file
Thilo StadelmannThilo Stadelmann
Thilo Stadelmann
JKJulian Keuzenkamp
Julian Keuzenkamp
HGHelmut Grabner
Helmut Grabner
CWChristoph Würsch
Christoph Würsch
We present the “AI-Atlas” didactic concept as a coherent set of best practices for teaching Artificial Intelligence (AI) and Machine Learning (ML) to a technical audience in tertiary education, and report on its implementation and evaluation within a design-based research framework and two actual courses: an introduction to AI within the final year of an undergraduate computer science program, as well as an introduction to ML within an interdisciplinary graduate program in engineering. The concept was developed in reaction to the recent AI surge and corresponding demand for foundational teaching on the subject to a broad and diverse audience, with on-site teaching of small classes in mind and designed to build on the specific strengths in motivational public speaking of the lecturers. The research question and focus of our evaluation is to what extent the concept serves this purpose, specifically taking into account the necessary but unforeseen transfer to ongoing hybrid and fully online teaching since March 2020 due to the COVID-19 pandemic. Our contribution is two-fold: besides (i) presenting a general didactic concept for tertiary engineering education in AI and ML, ready for adoption, we (ii) draw conclusions from the comparison of qualitative student evaluations (n = 24–30) and quantitative exam results (n = 62–113) of two full semesters under pandemic conditions with the result of previous years (participants from Zurich, Switzerland). This yields specific recommendations for the adoption of any technical curriculum under flexible teaching conditions—be it on-site, hybrid, or online.

30 June 2025 in Sensors
Verified
Journal Article
Open access
PDF file
MGMichael Gottinger
Michael Gottinger
NNNicola Notari
Nicola Notari
SDSamuel Dutler
Samuel Dutler
SKSamuel Kranz
Samuel Kranz
RVRobin Vetsch
Robin Vetsch
GPGuido Piai
Guido Piai
State-of-the-art radar systems for the contactless monitoring of vital signs and respiratory diseases are typically based on single-channel continuous wave (CW) technology. This technique allows precise measurements of respiration patterns, periods of movement, and heart rate. Major practical problems arise as CW systems suffer from signal cancellation due to destructive interference, limited overall functionality, and a possibility of low signal quality over longer periods. This work introduces a sophisticated multiple-input multiple-output (MIMO) solution that captures a radar image to estimate the sleep pose and position of a person (first step) and determine key vital parameters (second step). The first step is enabled by processing radar data with a forked convolutional neural network, which is trained with reference data captured by a time-of-flight depth camera. Key vital parameters that can be measured in the second step are respiration rate, asynchronous respiratory movement of chest and abdomen and limb movements. The developed algorithms were tested through experiments. The achieved mean absolute error (MAE) for the locations of the xiphoid and navel was less than 5 cm and the categorical accuracy of pose classification and limb movement detection was better than 90% and 98.6%, respectively. The MAE of the breathing rate was measured between 0.06 and 0.8 cycles per minute.

4 December 2023
Verified
Conference Paper
Closed access
PTPhilipp Gérard Trémuel
Philipp Gérard Trémuel
EGEfstratios Gavves
Efstratios Gavves
CWChristoph Würsch
Christoph Würsch
KFKlaus Frick
Klaus Frick
RVRobin Vetsch
Robin Vetsch
This paper presents a novel method for resolution free, free-form shape optimization for nonuniform transmission lines using artificial neural networks. We use a low dimensional representation of the geometries by learning a neural field embedding with a contrastive loss function to group similar geometries. A second neural network predicts the scattering parameters from the encoded geometry for a given frequency point, that are used to define the optimization objective in the frequency domain. The whole pipeline is fully-differentiable enabling the use of fast gradient based optimization methods. The proposed model architecture shows promising results for the simple test case of optimizing a transmission line taper. The method is fast, stable and flexible to apply to different geometries with different constraints and requirements.

14 July 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
RVRobin Vetsch
Robin Vetsch
LFLucas Falch
Lucas Falch
CWChristoph Würsch
Christoph Würsch
SGSebastien Gros
Sebastien Gros
Construction increasingly adopts automation and operator-assistance systems for safer, more productive machinery. Mobile-walking excavators operate in steep, uneven terrain but omit the rear counterweight of standard excavators, so stability depends on the undercarriage configuration and uppercarriage motion, leaving them prone to operator-induced tip-over. This work presents a Model Predictive Safety Filter (MPSF) that automates stability supervision for a 15-ton, 19-degree-of-freedom (DoF) KAISER S12 mobile-walking excavator. The filter passes operator joystick commands unchanged while safe and intervenes minimally, without reversing the operator's intent, when a stability violation is predicted. Stability is quantified by the Zero Moment Point (ZMP) margin with measured terrain inclination, solved with acados over a 2.5s horizon. In ten field tests, the three MPSF-inactive runs reached negative margins down to -0.145m, while the seven MPSF-active runs stayed at or above +0.071m without slowing the operation. Such tip-over prevention will be essential for future teleoperation and increasing autonomy.

13 June 2026 in ACM Transactions on Design Automation of Electronic Systems
Verified
Journal Article
Open access
PTPhilipp Gérard Trémuel
Philipp Gérard Trémuel
FBFernando Benites
Fernando Benites
CWChristoph Würsch
Christoph Würsch
UHUlf Hügel
Ulf Hügel
FWFlorian Wolf
Florian Wolf
EGEfstratios Gavves
Efstratios Gavves
Addressing inverse design problems in engineering generally requires managing complex challenges. Traditionally, such systems are optimized by manually setting up a model in a numerical simulation software and applying parameter optimization. Topology changes and variable parameter sets result in challenging mixed-integer optimization problems. Such problems are hard to solve automatically; that’s why they normally need to be initialized manually with high effort by the design engineer. Generative probabilistic models, like Large Language Models (LLMs), can efficiently produce diverse, valid design concepts that are fully parameterizable, simulatable, and optimizable. We introduce cGenEDA, a Co-Pilot –a system based on multiple LLM agents using in-context prompting to coordinate these tasks, providing a seamless, adaptable toolchain for automated design generation, optimization problem definition, and design automation. Our approach demonstrates that long, interacting simulation model code sequences can be generated accurately and directly used as model input for domain-specific simulators for precise forward simulations. Simultaneously, cGenEDA will set up the optimization problem, combine it with well-known, black-box optimization algorithms, and efficiently solve inverse design problems.