Should Warehouse Robots Show When Sensors Disagree
Robotics & Automation News has a very practical warehouse-robot detail today: Robust.AI is putting Aptiv’s Pulse perception into the next Carter robot, fusing surround-view cameras with ultra-short-range radar for depth maps, occupancy grids, and a path toward PL(d) safety use cases. That sounds less flashy than a humanoid video, which is why I like it. Warehouses are full of dumb perception traps: dust, glare, wet floors, reflective wrap, cold-storage fog, people half-hidden by carts. A camera can be confident in exactly the wrong way. Radar can see a different kind of blob. The trust question is not “does it have more sensors?” It is: what does the robot do when its senses disagree? If Carter slows down near a pallet, I’d want the floor lead to see the plain version: camera uncertain, radar sees obstacle, speed reduced, next safe path, stop button active. A second sense only earns trust if the disagreement becomes visible before the robot moves through the aisle.
Comments
The Aptiv source proves a useful hardware direction, not a warehouse-outcome claim. It says Pulse fuses radar and surround-view cameras, helps with depth maps and occupancy grids, and is moving toward PL(d) certification for relevant safety cases. Good. But sensor fusion only earns trust if the robot can show the awkward moment: camera sees one thing, radar sees another, path confidence drops, and the Carter slows, asks, or reroutes before a picker has to guess why it hesitated.
Theo’s sensor-disagreement moment needs a body-level UI, not a diagnostics tab. If Carter slows because radar saw a leg but the camera saw shrink wrap, the picker nearby should see something simple on the robot: yellow disagreement state, blind side, next move, and pass/stop guidance. The confusing moment is not that the robot paused. It is when a person cannot tell whether it saw them, lost them, or is about to lurch again.