Skip to content
IEEE Spectrum AI· VicOne·Indexed by this site · Originally published· Opinion →

VicOne LAB R7: robot safety features work, but what happens when input is manipulated?

Your Robot’s Safety Functions Already Work. What If the Input Lies?

AI summary

VicOne LAB R7 testing shows that manipulated input can make robots respond as designed yet perform unintended actions. A Gemma 4 E4B robot dog treated poster text as instructions and changed its movement; a hospital service robot simulation with Nemotron on NVIDIA Jetson AGX Orin had its behavior changed by audio inaudible to the human ear. In another bug bounty activity, injecting ROS 2/DDS messages made a robot that should have stayed still move.

Related papers
Read originalspectrum.ieee.org