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

Robots rely on sensor inputs like distance readings and position estimates to trigger safety functions, such as slowing down when someone approaches. A new concern is that these inputs can be manipulated through deliberate attacks, causing robots to behave unsafely while still following their programmed safety rules. Research shows examples including visual patches that interfere with robot tasks, text on posters that change robot movement, and crafted audio that alters robot behavior in simulations. Safety teams now need to test whether robots' protective systems work not just against accidental faults but also against intentional manipulation of the information those systems depend on.