How to build fault screens that operators actually see
TOS Robotics builds fault screens around a photo of the machine with the part to fix outlined in red and one short instruction, because operators read the screen from a distance and come to it after the line has gone quiet. We rebuilt our screens this way in July 2026 on a bell pepper line in the Netherlands, where not all operators read English. We have no measured before and after figure for how fast operators react.
First setup
Our first robot interface was made for developers and demos, with errors in pop-up dialog boxes. In May 2026 we built the first interface for production. It runs as a kiosk, a locked full-screen mode that the operator cannot close. Operators get one Start and one Stop for the whole line and two Empty buttons, and technicians get the rest in a separate view. If the interface stops reporting for 15 seconds, the robots stop.
What went wrong
Operators read the screen from a distance, not all of them read English, and they came to the screen after the line had gone quiet. One stop message told them to clean out the machine, while the logs of the next days showed that a reset alone cleared nearly all of those stops within seconds to two minutes. Each needless stop costs output, see why small errors cause big losses on a packing line.
A translation lost the words left and right, so the operator no longer knew which chute to clear. A single jammed flap raised up to three banners and turned both robots red, and the second banner sent the operator to the machine that was fine.
What we changed
In July 2026 we rewrote the messages with the customer as short commands and rebuilt the operator screen around pictures.
| Before | Now |
|---|---|
| Error text in a dialog box | A red banner with a photo of the machine, the part to fix outlined |
| Long or generic messages | One short command, such as “Remove stuck peppers from the V-belt” |
| English only | Four languages, text size locked to large |
| “Setup 1” and “setup 2” | “Right robot” and “left robot”, as seen from the front |
| Up to three alarms for one jammed flap | One banner with the cause, the rest in the technician log |
| The newest or oldest alarm on top | The most specific alarm on top |
Since October 2026 the outline on the photo pulses red, to be visible from a distance. A jammed flap outlines the flap itself, not the whole chute. A help button opens a step-by-step photo guide, and the paper manual is generated from the same annotated photos.
Red is for errors and the Stop button only, warnings are orange. A robot that stopped because of the other robot reads “Stopped”, not “Error”. An alarm clears on every screen when the fault clears or after a successful Start. Alarms that stop the run are also emailed to the TOS team.
Results
The operator now sees one banner per event, with the cause, the place and the side. The cleaning message became three stop messages, one per action. Getting the emails right took three rounds. We muted routine “Press Reset” mails after dozens in two weeks, fixed a bug that muted a right hopper alarm after a left one, and later mailed every error again to count how often each one occurs.
Questions and answers
What if nobody is near the screen?
Alarms that stop the run are emailed to the TOS team, and a controller that goes silent for 30 seconds raises its own alarm. A daily email at 09:00 has the cycle count and the mispick rate in its subject line, so it does not need to be opened.
Can you measure whether operators react faster?
The software logs the time from each alarm to the next Start press, so a before and after comparison can be made from the line data. We have not made it yet.