ROBBRIEF
VicOne turned its DEF CON 34 robot hacking challenges into a free NVIDIA Isaac Sim robotics simulator extension, the kind of open adversarial test platform that lets capable robots share space with people.
A robot dog on stage, a poster taped to a wall, and a safety rule that was technically satisfied while the robot steered the wrong way. That scene is driving the new frontier of robot safety research, and VicOne has put a public test bench for it on the shelf.
The company has packaged the robot-hacking challenges it ran at DEF CON 34 into a free NVIDIA Isaac Sim extension called LAB R7. The repository documents clean-baseline and adversarial-sample stations, captures camera frames, model actions and attention maps, and lists traffic-sign, ISO-warning, typography and adversarial-patch categories. The default out-of-the-box model is google/gemma-4-e2b-it, a small vision-language model; VLM action tests run immediately, while VLA (vision-language-action) trajectory tests need vendor contact. Running the lab asks for Isaac Sim 5.0 through 6.0, Ubuntu 22.04 or 24.04, and an NVIDIA RTX 3080 or better.
In a sponsored Robot Report piece, VicOne describes a ROS 2 (Robot Operating System 2) message injection at a bug-bounty event that nudged a robot expected to stay put, alongside a separate inaudible-audio hospital-robot simulation. These are vendor-reported demonstrations, and the article notes that an independent protective system can still interrupt the chain.
What remains unknown is whether the extension's scenarios translate into safety-certification gains. The release dates to August 9. The category is the news: adversarial-input research is now a downloadable test bench, and the robots that will live near us are the ones whose perception gets torn apart in simulation first.