01
Deciding where to observe
AME
Use the uncertainty in an attention map to choose informative observations.
↗Doctoral research
Learning to look and search
Efficient visual exploration for embodied agents. A research program about choosing useful observations and actions when sensing and computation are limited.
Download thesis PDF ↓Doctoral thesis · 2026
- Institution
- Jagiellonian University
- Supervisor
- Prof. Bartosz Zieliński
- Status
- Submitted in 2026
Where should an agent look next?
Many visual systems process a fixed image all at once. An embodied agent has another choice: it can move, look again, and spend its attention where it matters. My doctoral work explores that choice across several scales, from the next image region to a path through an environment.
The projects below form a conceptual sequence. Their publication dates do not necessarily follow this order.
Four connected questions
A progression of ideas
02
Seeing beyond fixed patches
Beyond Grids
Represent visual inputs when patch positions, scales, and coverage can vary.
↗03
Learning to take a glimpse
AdaGlimpse
Learn to select the position and scale of each observation.
↗04
Searching through an environment
FlySearch
Study sustained, goal directed exploration in realistic outdoor environments.
↗The complete work
Read the thesis
The thesis brings these contributions together and discusses the path from local visual decisions to longer horizon exploration.
Download PDF ↓Related publications