Chalmers Open Digital Repository
Välkommen till Chalmers öppna digitala arkiv!
Här hittar du:
- Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
- Digitala specialsamlingar, som t ex Chalmers modellkammare
- Utvalda projektrapporter
Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se
Enheter i Chalmers ODR
Välj en enhet för att se alla samlingar.
Senast inlagda
Utveckling av en AI-assistent för stöd vid en blodgivares hälsodeklaration - Testning, verifiering och implementering av LLM-baserade användargränssnitt för en medicinsk domän
(2025) Olofsson, Tyra
Med bakgrund av att frivillig blodgivning är en viktig del av svensk sjukvård finns
det intresse för möjligheten att låta en potentiell donator fylla i sin obligatoriska hälsodeklaration
digitalt hemifrån. En EHD-prototyp har utvecklats tillsammans med
en integrerad AI-assistent med syfte och funktion att besvara användares medicinska
frågor angående blodgivning. AI-assistenten nyttjar RAG för hämtning av domänspecifik
information, som har extraherats från SweBa:s regelverk för blodgivning.
Korrektheten i AI-assistentens genererade svar har manuellt granskats med ett resultat
på 93% korrekthet för frågor relaterade till sjukdomar och 97% för frågor
relaterat till vaccinationer. Resultatet bedöms inte tillförlitligt nog att användas i
EHD på grund av risk för felaktiga svar samt de etiska och juridiska och implikationer
som kan uppstå om en användare agerar utifrån felaktig information.
Data-Driven Automated Reporting Solution for External Collaborations - LLM-driven KPI Definition
(2026) Juul, Jakob; Lorentzon, Marcus
This thesis presents a proof-of-concept, developed with AstraZeneca (AZ), that explores
automating progress reporting for external collaborations by testing whether
a large language model (LLM)-driven system can extract objectives from contracts
and translate them into tailor-made key performance indicators (KPIs). Objective
extraction is quite reliable, reaching several highs of accuracy around the 85%-mark,
but converting objectives into KPIs that stakeholders judge as relevant, clear, actionable,
and measurable, is substantially less solid. Fewer than half of the KPIs
met each quality criterion on average, and 39% met none. Survey responses noted
that KPIs were often unclear, overly generic, or poorly timed, and skewed toward
simple counts (e.g., “number of models”) that miss quality and impact.
From interviews conducted at AZ, a set of general KPIs, that were deemed meaningful
to measure in a collaboration project, could be demonstrated. The final evaluation
suggests that these KPIs (e.g., external engagement and budget coherence)
outperform collaboration-specific KPIs generated directly from objectives. This underscores
the difficulty of creating bespoke target measures in diverse contexts.
Despite these issues, the approach offers practical value. In principle, the pipeline
should be better suited for agreements with explicit milestones (e.g., business or
commercialisation contracts), where more clearly defined expected outcomes support
better-formed KPIs. However, this cannot be conclusively established by the
implementation in this thesis, due to limited data.
Ultimately, translating qualitative objectives into quantitative, decision-grade KPIs
remains inherently difficult. Contemporary LLMs are capable across many aspects of
automation, but evidently less reliable for high-judgement and context-specific KPI
design that balances relevance, clarity, actionability, and measurability, at least by
following the approach outlined in this thesis. Therefore, the most defensible nearterm
usefulness is in metadata extraction and recommendation, while still requiring
a human-in-the-loop as a safeguard. In turn, this can improve customer relationship
management (CRM) metadata completeness and enable collaboration health
insights and automated reporting.
Reducing downlink reference signal overhead for CSI acquisition in massive MIMO systems
(2026) Führ, Andreas
The number of antennas used in massive multiple input multiple output (MIMO) systems
is expected to increase significantly to meet requirements of future radio access networks
(RANs). In legacy 5G New Radio, scaling the number of antennas imposes a proportional
increase in overhead associated with the acquisition of channel state information (CSI)
through downlink (DL) transmission of reference signals (CSI-RS). This thesis investigates
methods to reduce the overhead of CSI-RS transmission using a twofold approach:
optimising pilot placement for sparse sounding of the reference signals, and reconstructing
the full channel information from these sparse measurements at the user equipment (UE).
First, the sparse sampling of CSI-RS is formulated as a submodular optimisation
problem, and a cost function is presented based on the frame potential of the DL channel
estimated by the UE. A greedy algorithm for solving the sparse pilot placement problem is
proposed and evaluated for a simulated 3rd Generation Partnership Project (3GPP) MIMO
urban microcell environment with a uniform planar array (UPA), achieving near-optimal
pilot placement for subsets of antenna ports.
The second part of the thesis investigates the recovery of the full channel information
from the sparsely sounded CSI-RS using an artificial neural network (ANN). A
physics-informed U-Net architecture is developed, that leverages the sparse angular
representation of the DL channel to recover the full-rank channel. The ANN is trained on
a large dataset of simulated noiseless DL channels for the same 3GPP environment and
for several different spatial pilot configurations and muting levels.
The results of the experiments show that the ANN model can achieve a reconstruction
accuracy comparable to basis pursuit denoising (BPDN), while outperforming BPDN
in computational efficiency. In addition, the choice of spatial antenna port muting
pattern has a noticeable impact on the reconstruction performance of both methods in
the considered scenario, with the found near-optimal sampling patterns gives the closest
spectral similarity to the full-rank channel in terms of the Itakura-Saito distance. The
combined approach of optimising sparse pilot placement and using a neural network
for reconstruction demonstrates the potential of AI functionality for CSI-RS overhead
reduction and for improving the performance of massive MIMO systems in upcoming 6G
networks and beyond.
Transformer-Based Crystal Structure Generation from OTC and Chemical Composition
(2026) Peter, Anu
Multi-component oxides, composed of three or more elements, offer a vast combinatorial
space of possible structures with tunable properties such as thermal stability,
ion conductivity, and catalytic activity. Exploring this space using traditional trialand-
error methods is time-consuming and expensive.
This thesis investigates the use of a Transformer-based language model to generate
Crystallographic Information Files (CIFs), which encode atomic positions, lattice
parameters, and symmetry elements. The model is trained to learn relationships
between structural features and material properties, allowing it to propose new
CIFs representing potential novel crystal structures based on input descriptors like
oxygen transfer capacity and composition.
The results show that the Transformer model can capture complex structural patterns
and generate valid CIF sequences, demonstrating its potential as a data-driven
tool to accelerate the discovery and design of multi-component oxides.
Development of a Driver-in-the-Loop Advanced Driver Assistance Systems Prototyping Platform
(2025) Chouhan, Abhay; Fäldt, Linus; Jonsson, Elliot; Liao, Fanxiang; Sridharraju, Prakash Raju; Wang, Jingyu
This project presents the development of a small scale Driver in the Loop Advanced Driver Assistance System prototyping platform based on a MentorPi robot vehicle and a modular Robot Operating System 2 software architecture. The platform integrates human driver inputs (steering wheel, pedals, and gear selection), onboard sensing (monocular camera and 360° LiDAR), and real time visualisation into a closed loop test setup for safe and repeatable indoor testing. Functionality is distributed across Robot Operating System 2 nodes for driver input handling, motion control, LiDAR processing, state aggregation, and live camera streaming with information overlays.
A simplified Automatic Emergency Braking function is implemented using an Enhanced Time to Collision, using a stopping distance trigger. Rather than relying on a fixed distance threshold, this approach estimates whether the vehicle can safely stop before reaching an obstacle by accounting for the speed and braking capability, making the intervention more representative of practical safety behaviour. Two operating modes were validated: a baseline browser based camera view and a virtual reality mode using a Meta Quest 3. In the VR configuration, the camera feed is accessed via an HTTP snapshot endpoint, while head tracking data are transmitted over User Datagram Protocol to control the pan–tilt camera.
Results show stable baseline teleoperation, with 20 consecutive laps completed without system restart, and successful execution of a combined VR driving–slalom, Automatic Emergency Braking scenario in four out of five runs. However, VR operation occasionally led to a paralyzed control state, indicating integration and stability limitations under increased system load. Beyond functional validation, the platform is intended to enable rapid prototyping and early stage evaluation of Active Safety and Advanced Driver Assistance Systems concepts. Its low cost, modular design, and safe indoor operation make it particularly suitable for pedagogical activities at Chalmers University of Technology, supporting hands on learning and experimentation in courses related to Active Safety and Driver in the Loop system development.