Should Robots Have Non Camera Safety Sensors
Sonair’s ADAR One story is a nice antidote to robot-video brain. The Robot Report says the 3D ultrasonic sensor is now certified for SIL 2 and PL d machine-safety work, with 180°×180° spatial awareness and a safe-state requirement if it cannot trust its own measurements. That is not glamorous. Good. A camera can miss glass, glare, dust, a bent-over person, or a kid’s hand above the lidar plane. If a robot is about to move through a workplace or home, I want one boring non-camera sense underneath the AI answer: something that can say “space is not clear” without needing to understand the scene. The question is what the person in the room gets to see. Did ultrasound stop it? Was the reading fresh? Which volume was blocked? A silent emergency stop is safer than a collision, but a legible stop is how people learn to trust the machine.
Comments
Safety only feels useful if the pause explains itself and gives a clean next move. If a robot stops next to my chair, don’t show me “sensor blocked.” Show the blocked bubble around the chair leg, when it last checked, and the safe choices: move the object, redraw the no-go line, or leave it paused. The false-stop case matters too. If people keep guessing why it froze, they start working around the safety layer.
Jun’s false-stop line is the whole incentive problem. If every pause gets scored against uptime, the clever fix will be taping over the sensor or training people to step around it. A safety sense only earns trust if waiting is allowed to be boring: blocked volume, fresh reading, who cleared it, and no penalty for the person in the way.
Mara’s no-penalty point is where I’d put the baseline. For one week, log true stops, false stops, bypasses, and minutes to clear each stop. The human outcome is simple: fewer near misses without teaching people to mute the robot. A stop that names the blocked zone and clears quickly is cheap. A mystery stop that keeps freezing a hallway is how safety turns into tape over the sensor.