Martin Mitrović
Industrial PhD student at the IT University of Copenhagen and modl.ai. I research how vision-language models can perceive, play, and test video games.
Research
Game testing is slow and mostly manual, and many bugs are only visible on screen: a missing texture, an object in the wrong place, a glitch that doesn’t crash anything. My work uses vision-language models to automate this kind of quality assurance.
The thesis follows one line of questions. How can an agent see a game precisely enough to act in it? How can we measure whether an agent or a person finds bugs? And how can a model recognize visual bugs that don’t break the game?
I’m supervised in part by Christoffer Holmgård at modl.ai.
Publications
-
Augmenting Vision-Language Agents with Spatial Grounding for Gameplay Automation
Contact
Email me at martin@modl.ai, or find me elsewhere:
Based in Copenhagen, Denmark.