Here is a made-up tale that robotics operations folks will find uncomfortably familiar. A small warehouse team ran a handful of automated guided vehicles, and every so often one would stop dead at the mouth of aisle seven, display “path blocked,” and wait for a human. The aisle was clear every time anyone checked. The team lost weeks to it, until someone suggested the oldest trick going: explain the stoppage aloud to the empty charging dock, from the top, and see what falls out.
Weeks of theories
The early guesses were all reasonable. Maybe the lidar was dirty. Maybe a reflective floor patch confused the sensor. Maybe the traffic controller was holding the robot for a conflict that did not exist. Each theory got a ticket, and each ticket got closed with “could not reproduce.” The stoppages came about twice a week, never on a schedule anyone could spot, always at the same aisle, and always with nothing in the way.
The team’s supervisor, tired of status meetings that repeated the same facts, set a rule for one afternoon: nobody opens a dashboard. Instead, one technician narrates a complete robot journey aloud, from task assignment to the stop, as if the dock could hear. The others are free to listen but not to talk.
The narration
The technician began, a little self-consciously. “The robot gets a pick task from the fleet manager. It plans a route along the main corridor. It reaches the junction for aisle seven. It slows down. The scanner looks into the aisle. If nothing is in the safety zone, it turns in and carries on. If something is, it stops and reports ‘path blocked.'”
Then the technician said: “The zone the scanner checks… is defined where? On the robot. In its configuration. The size depends on the load it is carrying. A loaded robot has a bigger zone because it needs more stopping distance.” A pause. “And the stoppages happen when the robot is carrying the tall pallets. Those are the ones that stick out at the top. Does the zone follow the load’s width, or its height?”
The room went quiet. Nobody had linked the stoppages to which pallets were being carried, because the task logs did not show pallet type on the same screen as the stop events. The question, once spoken, was easy to check.
The check
The team pulled a month of stop events and matched them to the loads. Every stop involved a tall pallet. A look at the configuration showed that the safety zone was sized for the width of the load, with a generous margin, but the aisle seven entrance had a low-hanging cable tray, installed after the original layout survey. The scanning plane saw the underside of the tray as an obstacle only when a tall load raised the robot’s reference profile into its range. Empty aisle, real obstacle, as far as the robot was concerned.
Nothing in the data was missing. The story had simply never been told in order, from the robot’s point of view, with each step stated plainly. The dashboards showed events; the narration showed a sequence.
What they changed
- The cable tray was lifted slightly and the aisle entrance re-surveyed.
- The stop-event log now shows the load type next to the reason code, so a pattern is visible at a glance.
- The commissioning checklist gained a line asking, “Has anything been added overhead or at floor level since the last survey?”
- Any recurring “cannot reproduce” ticket now gets a ten-minute spoken walk-through before it is closed.
What the story teaches robotics teams
Robots make a particular kind of problem: the machine’s view of the world is different from yours, and it is built from sensors, thresholds and maps that nobody reads aloud. When something goes wrong, you are tempted to inspect the world, but the useful move is often to restate the robot’s world. Say what it sees, what it decides, and why, in order. For background on the sensing side, see the piece covering the way industrial robots incorporate their sensors, and for the vehicle context, a complete guide to AGV warehouse automation sets out how these fleets are organised.
There is a further point about teamwork, which this tale only hints at: the narrator was alone with the dock, in the sense that nobody could interject. The listeners did not help the technician reach the question. The discipline of finishing each sentence did.
Try it on your own fleet
A short script for the next stubborn robot fault:
- Start at the moment the task was assigned.
- Say, in order, what the robot knows, what it senses and what it decides.
- At every “it decides,” say which setting or threshold it uses and where it lives.
- Whenever you say “always” or “never,” ask whether the logs actually show that.
- Write down the sentence where you slowed down.
Reading outside the field
Finally, a small aside for people who like to bring good habits to unrelated sites. When a visitor opens GastroHospitals, a directory-style website about gastroenterology with hospital and doctor areas, a blog and a forum, it helps to state what you are after before picking a menu entry. The robot team did the same with its own system, only with higher stakes and louder alarms.
Last word from the dock
The fictional team never did find a faulty sensor, because there was not one. They found an unspoken assumption sitting between a configuration file, a pallet, and a cable tray. If your own fleet has a fault that “cannot be reproduced,” tell its story out loud from start to finish, and listen for the sentence that comes out slower than the rest.