Robotics and Semantic Systems

Computer Science | LTH | Lund University

Denna sida på svenska This page in English


CS MSc Thesis Presentation Day August 24 2023


From: 2023-08-24 09:15 to 16:00
Place: See information for each presentation
Contact: birger [dot] swahn [at] cs [dot] lth [dot] se
Save event to your calendar

Five MSc theses to be presented on Thursday August 24, 2023

Thursday August 24 is a day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Five MSc theses will be presented.

You will find information about how to follow along under each presentation. There will be presentations in three different rooms: E:2405 (Glasburen), E:2116 and E:4130 (Lucas). A preliminary schedule follows.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation ( Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.


E:2405 (Glasburen)



Presenters: Rune Anderberg, Henrik Olsson
Title: Turning Disengagement Reports Into Executable Test Scenarios for Autonomous Vehicles Using NLP
Examiner: Per Runeson
Supervisor: Qunying Song (LTH)

Autonomous vehicle development is rapidly growing, and occasionally, testing occurs on public roads. Although valuable, such testing can also be expensive and irregular. To be allowed to test on public roads in California, USA, manufacturers must submit reports containing each instance the autonomous driving mode had been disengaged, either by the driver or the vehicle, to the California DMW. This master’s thesis has studied using these disengagement reports as a basis when creating testing scenarios for simulators to ease and lower the cost of testing to increase the safety of autonomous vehicles. We developed a concept application that automatically creates test scenarios from disengagement reports using Natural Language Processing. Due to issues with the disengagement reports, the application can only create concrete test scenarios for a small subset of the disengagement reports. However, the project demonstrates the feasibility of this approach and proposes future work through suggested improvements and ideas.

Link to popular science summary:


Presenter: Elias Sjöberg
Title: Generating Synthetic Scenarios to Test an AI-Enabled Traffic Measurement System
Examiner: Per Runeson
Supervisor: Markus Borg (LTH)

To make the future of the transport system efficient, safe and sustainable, understanding of the complexities of traffic is vital. Viscando provides detailed traffic data by utilizing Machine Learning and computer vision algorithms in their infrastructure sensor OTUS3D. The purpose of this study was to test OTUS3D using simulated traffic scenarios. Using the MathWorks package RoadRunner, we created a digital model based on a real-life junction. This digital model was imported into the open-source traffic simulator CARLA. We developed two models to generate parameter values that were used to create traffic scenarios: one of the models used random sampling while the other was based on the genetic algorithm NSGA-II. Due to time limitations, it was not possible to create a sufficient number of simulated scenarios to properly evaluate the two models. However, some conclusions could be drawn regarding which traffic scenarios were particularly challenging for OTUS3D to estimate.

Link to popular science summary:




10:15-11:00 N.B. Change of room

Presenters: Joakim Mörling, Joseph Atalla
Title: Investigating Optimization Techniques for Algorithms in Radars
Examiner: Per Andersson
Supervisors: Jonas Skeppstedt (LTH), Aras Papadelis (Axis Communications AB), Henrik Lehtonen (Axis Communications AB), Santhosh Nadig (Axis Communications AB)

This thesis analyzes bottlenecks in the radar processing chain at Axis Commu- nications AB, and attempts to investigate various optimization techniques for such bottlenecks. One bottleneck in particular was deemed bigger than the oth- ers, and is the focus of this thesis. The bottleneck is a clustering algorithm called DBSCAN. When more input is introduced than the algorithm can handle, then time available for other processes is reduced. A currently implemented down-sampling solution aims to alleviate this is- sue by decreasing the input size, thereby guaranteeing consistent performance. However, this approach has the undesirable side effect of reducing the radar’s clustering accuracy. In response, this thesis investigates alternative solutions that directly address the core problem, with the goal of higher performance. Some dif- ferent algorithms were tested, and a new algorithm was found that can increase the throughput by over 30 times. Attributes of different clustering algorithms have also been mapped.

Link to popular science summary:


Presenter: Fritjof Bengtsson​​​​​​​
Title: Facilitating the Development and Release of Research Software into an Open-Source Project​​​​​​​
Examiner: Martin Höst​​​​​​​
Supervisors: Alma Oruceviv-Alagic (LTH), Alexander Ekman (LU)

This work presents an account of the process of transforming Baler, a data compression tool for scientific data, into an open-source software project. Initially created for specialized scientific use, the expansion into an open-source project required combining the expertise of various scientific fields. Navigating this process presented several challenges, including managing a steep learning curve for scientists new to the industry standard of software development, deciding on a license, and addressing technical issues with limited existing solutions. The transition of Baler highlights the potential of open-source approaches in scientific software development. The interdisciplinary collaboration established during the process marks the value of combining diverse knowledge and skills. This experience provides valuable insights for similar projects and emphasizes the importance of open-source development in broadening the accessibility of scientific software tools.

Link to popular science summary:


E:4130 (Lucas)



Presenters: Oskar Larsson, Ludwig Jakobsson
Title: Camera Calibration for Alignment of a Real World Setup to a Simulated Environment​​​​​​​
Examiner: Elin A. Topp​​​​​​​
Supervisor: Alexander Dürr (LTH)

This thesis investigates camera calibration for use in computer vision systems. The intended use case is camera alignment between a simulated environment and a real world setup to facilitate sim2real learning. This is a two step process, intrinsic and extrinsic calibration. In both cases, the effects of several factors are investigated. These include image/pose count, rotational variation and distance. Included is also a proposal for a method to validate the calibration results in the form of a pick and place task with HSV object detection. The camera calibration is performed using a ChArUco calibration board, a camera attached to the robot's end-effector and two external cameras, one ceiling-mounted and one in front of the robot, facing the workbench.

Link to popular science summary: