Science in Luxembourg: AI-assisted drones to 1960s pop culture - inside the 2025 FNR Awards
Five awards at this year's FNR Awards underlined a simple truth: good science is both precise and collaborative. Three winners stood out for their blend of technical depth and public value - Florian Felten for an Outstanding PhD Thesis, and Professors Andreas Fickers and Sonja Kmec for Outstanding Scientific Achievement.
Multi-objective decisions that match real human trade-offs
Florian Felten's thesis on multi-objective reinforcement learning (MORL) grew from work at the University of Luxembourg's Interdisciplinary Centre for Security, Reliability and Trust (SnT). The team explored software that can generate other software, then applied it to coordinate drones and satellites acting in teams.
His core point is practical: decisions have competing goals. Time vs. cost. Speed vs. safety. Convenience vs. emissions. Most systems still optimize a single metric. Felten built algorithms that model user-specific preferences directly, so recommendations reflect what a person actually values - not the population average.
Beyond routing advice, the tech shows up in cooperative drone defense, where a swarm identifies and removes intruder drones. Other teams are already extending the work into water management and epidemic deconfinement scenarios.
Felten is clear-eyed about constraints. Compute scarcity limits experiments and risks concentrating know-how in a few firms. AI is a tool that can make mistakes. And hype invites misuse: people apply AI where it adds little value - a fast way to burn trust and budgets.
On a human level, the award validates years of uncertain progress. Professionally, it matters in an academic market where proof of contribution can open doors.
- For AI teams: design objectives as vectors, not a single score. Capture preference signals early. Use off-policy evaluation to compare trade-offs safely.
- For lab leads: model compute as a project constraint from day one. Plan partnerships without handing over the crown jewels. Keep a "does this need AI?" gate in your review process.
Learn more about SnT's work at the University of Luxembourg here, and about Luxembourg's research ecosystem via the FNR here.
Reframing the long 1960s: pop culture as a transnational network
Professors Andreas Fickers and Sonja Kmec led the "Popkult60" project from 2018 to 2025, a multidisciplinary look at European popular culture in the long 1960s. Phase one covered the obvious pillars - commercial radio like Radio Luxembourg, television, and music - using a classic historical approach.
Phase two went hands-on. The team ran experimental studies around events such as the Schueberfouer, and even reenacted border-region dance parties from the 1950s-60s to study practices in context. This isn't typical for historians, but it produced evidence you can't get from texts alone.
Findings challenged common assumptions. The 1960s weren't a one-way "Americanisation." Cultural flows crisscrossed Europe and even fed back to the US. One example: French and Belgian comic albums evolved into respected formats for adults, influencing the American market in the 1970s-80s. Tourism research showed similar multi-directional transfers once Indian Ocean countries gained independence.
The project also tested new ways to share results: a virtual exhibition with radio recordings, amateur films, and youth magazine covers, plus a book due next year with Transcript Publishing. The award recognizes method and team culture as much as outputs - and marks the first time the scientific achievement award went to the humanities.
- For humanities researchers: prototype methods (reenactment, participatory observation) to study practice, not just narratives.
- For consortia: make "transnational" a property of each case study, then assemble the mosaic. Pair classical analysis with public-facing formats to broaden impact.
What this means for research professionals
- Preference-aware AI will set the bar for human-aligned systems in policy, mobility, and environmental management. Plan for multi-objective evaluation now.
- Method innovation isn't limited to STEM. Experimental humanities can surface evidence traditional archives miss - and connect better with the public.
- Awards matter, but process matters more: open collaboration, fair compute access, and honest problem selection beat hype every time.
If your team is building applied AI and needs structured upskilling, you can browse focused programs by role here.
Your membership also unlocks:
