We're attending the 11th International VDI Conference – Connected Off-Highway Machines (April 29–30), and we're not just there to listen.
Defense autonomy doesn't get the luxury of clean roads or predictable scenarios. Autonomous systems operating in the field face novel obstacles, unfamiliar terrain, and long-tail edge cases as the norm rather than the exception. Traditional perception systems, however capable, were never built for this.
“Vision–language models are changing what autonomous defense vehicles can actually see, and understand.”
That's exactly the problem Embedl is helping to solve. On Wednesday, April 29th at 16:45, our own Peter Kristiansen takes the stage at the VDI Wissensforum GmbH conference to explore how vision–language models (VLMs) are redefining scene understanding for autonomous platforms operating in the most demanding environments imaginable.
What the talk covers
- How VLMs enable language-grounded scene understanding for autonomous defense vehicles, going beyond bounding boxes to contextual reasoning about what's actually out there.
- This leads to improved handling of novel obstacles, off-road terrain, and edge cases that classical perception systems weren't designed to cope with.
- What it actually takes to deploy VLMs on embedded, automotive-grade hardware, the real engineering constraints, and how Embedl tackles them.
The defense context is uniquely unforgiving: there are no second chances when a system fails to recognize an obstacle or misreads a scene. VLMs offer a level of semantic flexibility and out-of-distribution robustness that rule-based and purely statistical approaches simply can't match, and deploying them at the edge on constrained hardware is where Embedl's core expertise lies.
If you're attending the conference and want to connect, reach out to Peter directly or schedule a meeting below.