Embodied AI and robot learning · Denmark
Rune
Harlyk
My bachelor thesis put reinforcement learning on top of a vision-language-action model and measured whether it helped. Mostly it did not. Working out why is the part I want to keep doing. Alongside that there is a quadruped and a hexapod, both running their control loops on an ESP32.
- Projects
- 8
- Active
- 2
- On hardware
- 3
Selected work
All 8 projects →Robust vision-language-action models
FinishedBachelor thesis: can reinforcement learning improve a VLA policy after fine-tuning?
Spot Micro · Leika
ActiveA quadruped on an ESP32-CAM, with a learned policy correcting the hand-written gait.
Hexapod
ActiveA locomotion policy trained in simulation, running on the microcontroller itself.
Ferrofluid Visualizer
FinishedAn alien friend that dances to your music.
Bunny OS
FinishedA desktop operating system UX living inside a browser tab.
Currently
Starting an MSc in Human-Centered Artificial Intelligence at DTU in September 2026, after finishing the BSc in Artificial Intelligence and Data there in June. Writing software at Edora A/S alongside it.
Most of my spare time goes to the quadruped and the hexapod. Right now that means gait tuning and the app that drives them.
Get in touch if any of it overlaps with what you are building.