What ELROB Is (and Isn’t)

ELROB — the European Land Robot Trial — is a biennial field robotics event that has been running since 2006. This year’s edition, held at the Waffenplatz in Thun, Switzerland, from June 15–19, was the 13th installment and, by participant count, the largest in the event’s 20-year history. 21 teams. Five scenarios. A military training area that, at roughly 6.5 square kilometers, is the oldest and largest of its kind in Switzerland.

It is worth saying clearly what this is not. It is not a trade show. The framing from ELROB’s initiator, Dr. Frank E. Schneider of Fraunhofer FKIE, has been consistent since day one: ELROB is not a competition in the DARPA Grand Challenge sense. There are no official podiums. The goal is to test what’s actually possible, expose what isn’t, and close the loop between military users, researchers, and industry in a direct and practical way. The 2026 edition was hosted by armasuisse Wissenschaft und Technologie, Switzerland’s Federal Office for Armament.

After the event, Fraunhofer FKIE summarised it with a quote that landed well:

“It was a big family meeting.”

Those words came from Dr. Schneider himself. Twenty years in, that description still fits.

📸 All event photography by Fabian Vogl / Fraunhofer FKIE

📸 And some photos credited to their owners.

Getting There

How does one actually register for ELROB? It’s simpler than you might imagine, but you can’t just show up and expect everything to be easy. Registration is handled through Fraunhofer FKIE, with a hard deadline of January 31st each year. Teams fill in a team application (including a liability statement and photograph/video release form), a vehicle specification sheet, and a team membership roster. Media and visitor registrations are separate tracks.

As a visitor, the key thing to know is that you have to register in advance. There is no walk-in gate. The event takes place on an active military base. But once in, the communication and hospitality from the organizing team — local lead Thomas Nussbaumer from armasuisse received several public mentions for exactly this — were warm and professional. The venue also hosted an accompanying exhibition on the opening Monday, where teams and exhibitors could showcase systems in a less-pressured environment.

Communication from the organizers before the event was clear and informative, with none of the noise that often surrounds large technology gatherings. Noise, as one of the people I spoke to noted, is not the motivation for this gathering.

What I Saw

The trial ground during preparation days was what I can only describe as an open-air co-working space for field robotics. Tents were set up like workstations. Between improvised workshops, pallets of water, and equipment crates, teams ran cables, debugged systems, and prepared for their time slots. All types of robots were present — quadruped, wheeled, tracked, palleted, small, medium, and large — and their handlers were connected to them, tethered or otherwise, practicing before the scenarios began.

Some faces I recognized from entirely different contexts — polished automotive conferences, clean booths, controlled demonstrations. Seeing those same people crouched over robots in a field, in the heat, with dust on their hands, was a different thing entirely. The scenery changes people. Or maybe it just reveals them.

Walking that trial ground — the drills, the infrastructure, the institutional weight of it — took me somewhere I hadn’t expected to go. ELROB is a civilian-scientific event, but it lives inside a military reality. That tension is not something you feel at trade shows.

The Scenarios and the Winners

Five challenges ran across the week. Here’s a brief sketch of each, followed by who came out on top.

Convoy asked teams to have an autonomous vehicle follow a crewed lead vehicle across challenging terrain. Rheinmetall’s Mission Master, powered by Rheinmetall’s PATH‑Autonomy Kit (PATH‑A), won this category. Their system had already taken first in the 2024 Convoy scenario in Trier — and repeated here. Their LinkedIn post was notably matter-of-fact:

“Strong performance — and a great reflection of the work behind the scenes.”

That’s the right register for ELROB.

Reconnaissance required teams to search the terrain and build detailed situational awareness, including detection of objects of interest.

The joint team of ELP GmbH, European Logistic Partners and Roboverse Reply took the win — and also received an award for greatest team effort, after lending their Persistent Systems MPU5 mesh radio to Team Capra when their communications failed.

ELP executive Alexander Maus’ post-event reflection was a very honest read: they won, and still used the moment to articulate what events like ELROB are actually for — the audience sees a functioning system and a result; the developers see every module that wasn’t finished, every assumption that got challenged, every gap that still needs closing. Both perspectives are true, he wrote, and that tension is exactly the point.

Mule — the most popular scenario this year, with the largest number of participating teams — involved training a robot on a route and then sending it on as many autonomous round-trips as possible within 40 minutes, including a final data upload.

Diehl Defence won. But the Mule stories were not just about the winner. Read more to find out.

CasEvac / Search & Rescue tested robots’ ability to locate and recover casualties from difficult environments.

Telerob won with their Telemax EVO PLUS. The week also reinforced something the team took seriously: military users were direct about what solutions they needed, and how fast they needed them.

ERW (Explosive Remnants of War) — no separate winner was announced publicly beyond the ELP/Roboverse win, covering both Reconnaissance and ERW jointly.

The Mule in Detail — What I Missed and What Others Reported

I had planned to attend the Mule scenario on Wednesday and Thursday, but couldn’t make it. So here is what I gathered from the teams who were there.

The setup is deceptively simple: the UGV first follows a human operator to memorize a route, then navigates it back independently, then runs it again autonomously — managing barbed wire, dead ends, and watercourses along the way. Teams get a 15-minute preparation window immediately before their run, and the exact mission parameters are disclosed only then.

Eleven teams entered this scenario, making it the most contested discipline of the week.

Diehl Defence won it — for the second consecutive time. Their ZIESEL UGV, fitted with the PLATON Autonomy Kit, outperformed ten other teams in Thun, having done the same in Trier in 2024. Back-to-back wins in the most popular ELROB scenario, against a growing field, are not a footnote.

But the field itself was the story.

Poland’s Łukasiewicz – PIAP brought their PIAP IBIS TRANSPORT platform to the Mule, implementing autonomous supply transport in demanding non-urban terrain.

Their Head of Security Systems, Jakub Główka, said it directly after the event: ELROB is primarily about learning how to quickly adapt technology in real, demanding conditions — and that is its greatest value.

Prof. Roland Scherwey’s Team ROSAS from HES-SO Fribourg, running the GARM II platform from RUAG, experienced a critical motor driver hardware failure before their run even began.

Half the team drove to Stuttgart overnight to source a replacement part. The other half stayed and prepared for an early-morning repair. Scherwey called it

“a bit like being part of a Formula 1 pit crew after an unexpected crash.”

They made it to the start line and completed most of the mission objectives.

Team Capra from École de technologie supérieure in Montréal — who had already had a rough week, with communication failures in the Search & Rescue round that required borrowed radio equipment from ELP just to re-enter — also competed in the Mule.

On the approximately 300-meter course, they didn’t complete all required tasks, but the jury noted their improvisational talent and team spirit. Étienne Le Guerrier, team boss, had said before the event that they had fundamentally revised the platform and were testing it under real conditions for the first time. That context matters when reading the results.

Rheinmetall’s video of Eleni Sabourin demonstrating the Mission Master SP’s follow-me capability through obstacles and tight terrain is worth watching directly. It’s a clean look at where that technology currently sits — capable, precise, and still guided by a skilled human in the loop.

Watch here > Paul Rocco (Rheinmetall)'s LinkedIn post

A Question I’m Sitting With: Where Are the Turkish Teams?

While browsing through past ELROB photos, I noticed that Turkish teams have been there before. At ELROB 2024 in Trier, ITU RAKE — the Robotic Search and Rescue team from Istanbul Technical University — registered for the Search & Rescue scenario. In 2026, in Thun, they weren’t in the participant list.

I’m genuinely curious why. Turkey has strong engineering universities, a growing defense robotics sector, and real institutional interest in unmanned systems. Is participation too expensive logistically for university teams traveling from outside Turkey? Is it an awareness and network gap — the word simply not traveling as easily? Something else entirely?

I don’t have the answer. But I’d like to find out.

This is also personal for me. At the company I co-founded once (now in the past), we sponsored Turkish autonomous systems student teams for years. Watching that community grow up close gave me a real appreciation for the talent and drive that exist there. If any Turkish team is looking to participate in events like ELROB and needs guidance, connections, or support to get there, I’d be genuinely glad to help.

ELROB and Its Sister Events

One thread worth pulling: ELROB and ENRICH are twin events that are run in an annual rotation by the same core team at Fraunhofer FKIE, led by Dr. Frank E. Schneider. While ELROB focuses on military field robotics in outdoor terrain, ENRICH (the European Robotics Hackathon) is held every two years at the Zwentendorf nuclear power plant in Austria — the only nuclear facility in the world built but never commissioned — and focuses on CBRNE scenarios: robots navigating radiation, mapping hazardous environments, and performing manipulation tasks in conditions no human should enter.

They are philosophically identical events: not competitions in the trophy sense, but performance comparisons that expose what works, what doesn’t, and what needs to be built. The next ENRICH is planned for 2027. If you are in this space and haven’t looked at it, you should.

And ELROB 2028? Austria. Dr. Schneider’s team is already thinking about it.

Closing

Twenty years ago, someone at a NATO workshop noticed a gap. Military robots existed in theory but failed badly in practice. A framework was proposed to bring users, researchers, and industry together regularly, in the field, under conditions that did not flatter anyone.

That framework is still running. In 2026, it drew 21 teams, including students from Montréal who flew 6,000 km, engineers who drove to Stuttgart at midnight for a motor driver, and a Boston Dynamics Spot that apparently performed flawlessly in 35°C heat and dust.

Looking forward to another edition of ELROB, already! Maybe I’ll bring a team to the next one 😏

ICRA was never off my radar. For years, while curating the Autoware Foundation’s event calendar, I tracked it closely — checking the program, scanning the workshops, hoping we could get Autoware-affiliated voices on stage to tell that story to the robotics research community.

It never quite happened — and honestly, that made sense. Autoware was purpose-built for Level 4 autonomous driving. That’s its embodiment, its community, its reason for being. ICRA serves a much broader robotics audience, and while the technology fundamentals overlap significantly — Autoware is a ROS-based project after all — the fit was never quite there. I should also say that within the Autoware Foundation community, there is a substantial and growing group of researchers working to advance the broader robotics domain. The interest is real. But for a community as established and as focused as Autoware’s, pivoting toward the wider robotics industry is a sharp transition — one that takes time to navigate, if it happens at all.

This year, my engagement with ICRA was entirely different. I attended in my growth and community role at Nomadic AI — and Nomadic’s connection to the conference was anything but loose. I advised the team on exhibiting at the expo, we secured one of the last remaining booths on the floor, and our designer dressed it beautifully. The event organization was supportive throughout. We spent five full days in Vienna and felt, genuinely, at home.

Two discomforts

Now, the part where I didn’t feel completely comfortable — and there were two distinct flavors of discomfort.

Discomfort one: the organizational shift

Before Nomadic AI, events have been an integral part of what I’ve done. I knew the domain well enough to find my footing without too much trouble — but that doesn’t mean it was effortless. The friction I experienced there was real: slow-moving decision cycles, optics management, too many meetings, blurred lines of ownership and responsibility. I had learned to navigate all of that, and I earned my freedom to operate.

Nomadic AI is structurally different. Almost no meetings. Everything asynchronous. Heavy use of AI tools across business, sales, and marketing operations. The team is young and moves fast — sometimes faster than I find comfortable. I’ll be honest: I like to think things through carefully, in detail, and at a pace that lets me do that well. Adapting to a different rhythm has been its own challenge. I don’t say that as a complaint — I say it because pretending otherwise wouldn’t be accurate.

Running an expo presence is not a passive activity. The concurrent demands are relentless — scoping the floor, following up on conversations before they went cold, setting up meetings in the margins of a packed schedule, making real-time decisions about where to be and who to talk to. All of that while keeping the booth alive and the right conversations happening in front of it.

Five days. A booth. A co-organized side event. A full conference and workshop program runs in parallel. The skill of quickly deciding what matters and what doesn’t is tested in real time under those conditions. I felt that acutely in Vienna — and I hope that I learned quite a lot in those five days.

Discomfort two: the domain gap

The second discomfort is harder to resolve quickly: the technical depth of applied AI and my affinity for it. I came from autonomous vehicles, assuming robotics would be a short hop — shared vocabulary, overlapping architectures. It turns out to be much more nuanced than that.

The domain gap, unpacked

In autonomous vehicles, the shift from modular pipelines toward end-to-end learned systems — VLAs, world models, foundation model-based stacks — is still very much a live debate, not a done deal. This is an industry that has been running for well over a decade now, sustained by enormous capital conviction even as profitability and real returns remain elusive. Billions have been deployed. Market leaders have largely emerged. Consolidation is underway. And yet the question of how to cross the gap from experimentation and PoC to genuine scaled deployment remains open — complicated further by the rise of emergent players in the US and Asia while Europe visibly lags behind.

Robotics is earlier in that same arc, but with a fundamentally different technical character. Learned systems aren’t optional there — they’re fundamental. You simply cannot hardcode every use case when a humanoid has 40+ degrees of freedom and needs to manipulate tools, turn knobs, and imitate the full dexterity of human work. The complexity is of a different order from that of a car navigating roads. This has to be learned, or it doesn’t get built. The funding conviction is there — much like AVs before it. When the returns will follow is the open question.

One thing I’ve discovered since joining Nomadic: this is the era of the researcher. PhDs have rarely been more sought after or more consequential. The intellectual entry barrier is high and will remain so — that’s the nature of the work. But the research tooling barrier is a different matter. Tools are becoming more widely available, and open source plays a real role in lowering the barrier for labs that lack the means to build everything from scratch.

To close my own domain gap, I started going through robotics courses that have been made freely available online. These are structured, rigorous courses from leading institutions — not YouTube tutorials. Stanford’s CS223A covers the mathematical and mechanical foundations of robotics. The MIT compilation curated by Mithi Sevim spans everything from control theory to computer vision. ETH Zürich’s Robot Learning course dives deep into learned systems directly relevant to what I’ve been encountering at Nomadic. If you’re in a similar position — coming from an adjacent field and trying to build real fluency — I’d recommend starting here:

• Stanford CS223A

• MIT Robotics Coursework

• ETH Zürich Robot Learning

What I observed at the expo and conference

The north star everyone at the conference is oriented toward: Humanoids. They were everywhere at the expo — robot hands, full humanoid platforms, impressive hardware on every other booth. But look more closely, and most of it is being sold to research labs or for data collection purposes, not to factories or homes. That’s not a criticism — it’s an honest read of where the field actually is. The research community knows it too; they’re refreshingly candid about it in the sessions and hallway conversations.

What gave me genuine relief was that nobody has it fully figured out. The healthy disagreement between camps, the proliferating architectures — VLAs, vision-force-action models and beyond — and the candor about unsolved problems. It reminded me of the early days of autonomous vehicles. Everyone senses the inflection point. Nobody is certain which path reaches it first.

One of the most active conversations on the floor was about data — specifically, training data collection. Teams from across the world, some backed by their governments, are building data stacks to offer for model training. The creativity in how people are approaching aggregation was genuinely impressive. The conversations at our booth kept circling around the same underlying question: raw data alone isn’t enough. How do you add an understanding layer on top of it? These are exactly the conversations worth having, and we intend to keep having them.

I should also be honest about something I observed beyond the expo floor — in the narratives circulating on LinkedIn and in some of the coverage from the conference. Some of what I read painted a picture of humanoids that didn’t match what I saw in the sessions and workshops. Deployment-ready narratives for technology that, by the research community’s own admission, is still years away from industrial settings, let alone homes. I don’t enjoy pointing this out, but I think it matters. Paid promotion dressed as field insight creates a distorted baseline — and that distortion has consequences for how resources get allocated and how expectations get set. The first real venue for humanoid deployment will be industrial settings. That’s the honest read, and I’ll hold to it.

The comparison to watch: Automate in Chicago is coming up. It will be a different audience, a different commercial context. I’ll be watching closely to see whether the narratives shift when the room is full of buyers rather than researchers.

The good stuff

Open source had a strong pulse at ICRA — and that, more than anything, sent me home in good spirits.

From the AV side, KE:SAI’s commitment and vigor on stage were exactly the energy I needed. Seeing colleagues from TUM FTM carry that same flag through the ICRA hallways made it better. On the robotics side, the open source momentum is currently concentrated on teleoperated data acquisition tooling — which I think is exactly the right place to start commoditizing, alongside baseline models. Franka Robotics open sourced their acquisition and behavior suite #LABS right after the conference. And AIRoA’s initiative — open data and open models under a non-profit structure (one I’m quite familiar with) — is a signal worth watching closely.

Then there were the people. Reconnecting with longtime colleagues, meeting so many new faces, and — unexpectedly — stumbling upon a group of around 50 Turkish robotics researchers I had no idea was there. I intend to stay in contact. Every unfamiliar conversation was worth having. It was great to see and meet you all.

The ROS Meetup was a personal highlight. As a co-organizer, I was handed a bag of drink tokens and promptly spilled them across the floor. It turns out world-class researchers can absolutely be lured with a handful of plastic — perhaps because they were called tokens. Thanks to Kat for doing the real organizing work behind the scenes, and to Intrinsic, Foxglove and Prefix.dev for making it happen together. Nomadic AI was proud to be part of it.

And a deserved honorable mention: the Roboracer Foundation competition. My first time watching live — I hope not my last. The hard work the organizers have put in, with limited resources, has produced something genuinely impressive. Kudos to Rahul, Venkat, Felix, Johannes, and everyone who has built it into what it is today. A special mention to Chinmay and Tanmay for the Roboracer Sim Racing League: 73 teams in the virtual race for the ICRA edition. That’s not a side event. That’s Premier League-level work.

ICRA 2027 will be in Seoul. I hope to be there!