Assurant OmniBug
A seven-bay debugging station for phone refurbishment.
Assurant refurbishes millions of phones a year, and every one of them passes through a debug bench. I studied how operators actually work those benches, then designed a seven-bay station where an overhead camera and AI agent handle the screen work, so operators only load and unload.

[01]
Context
Every refurbished phone passes through a debug bench before it can be resold.
Smartphone refurbishment
Diagnose + wipe step
Mt. Juliet, TN
Assurant buys back, refurbishes and resells phones by the millions. Before a phone can be graded, repaired or resold, it is plugged into a computer, put through functional tests, restored and certified wiped. That step happens at a debug bench, one operator and one batch of phones at a time. OmniBug is a new station for that step: seven bays under an overhead camera, where an AI agent does the screen work so the operator only loads and unloads.
259K
Sq ft refurbishment center
~800
People testing and refurbishing on site
22M
Devices Assurant repurposed in 2023
[02]
The problem
Every phone waits on the slowest phone in its batch, and on the operator reading its screen.
~10 min idle per batch
~45 s of attention per phone
~15 phones per hour
~10 min
Idle port time per batch
~45 s
Operator attention per phone
~15 / hr
Phones per operator, legacy bench
[03]
Field study
Operators were shadowed across a 259,000 sq ft refurbishment floor.
Contextual inquiry
Shift shadowing
Reach + posture mapping
Operator + lead interviews

259,000 sq ft · ~800 people
[Fig.] A high-volume electronics floor, for scale. Representative image, not Assurant’s facility.
[Q1] Research question
Where does an operator’s time go between plugging a phone in and routing it out?
[Q2] Research question
What makes one device wait on another when a batch is connected at once?
[Q3] Research question
How do reach, posture and screen reading strain build over a shift?
[04]
Legacy bench
Eight steps, and the operator carries every one of them.
One batch at a time
Mac mini, hub, monitor
Fully manual
[05]
Ergonomics
Two places to look, eight cables to pinch, and one hub that can stop the line.
~800 mm reach
~60° head turn
8 screens to watch
[06]
Key insights
Five things the floor taught us.
Continuous flow
Let the machine watch
Position is identity
How might we
Make debugging a continuous flow instead of a batch, take screen watching off the operator, and give every phone a place that tells the system what it is?
[07]
Requirements
Seven requirements, each traced back to the field study.
Traced to insights
Measurable targets
Scorecard for every concept
[08]
Existing solutions
Robots are autonomous. Benches are cheap. Nothing on the floor was both.
Robotic test cells
Multi-port benches
[09]
Ideation
Cabinets, kiosks and benches, sketched side by side.
With Andrés Belisario and Levi Joo

[Fig.] Early form exploration across cabinet, kiosk and bench archetypes
[10]
Concept directions
Three directions, scored against all seven requirements.
Open-front bench selected
7 of 7 met
[11]
Human factors
Sized for a 5th percentile woman through a 95th percentile man, standing or perched.
5th to 95th percentile
Deck at 1,195 mm
Display at 1,726 mm
[12]
System architecture
The camera closes the loop the operator used to close.
One overhead camera
AI agent drives the software
A QR code per tray
[13]
Design details
Every touchpoint does exactly one job.
Lift-out trays
QR identity
Tool-free service
[A] The tray
[B] The enclosure
[14]
Design for manufacture
Sheet metal, tube and off-the-shelf parts, so a fleet can ship without hard tooling.
Low-volume processes
No hard tooling
[15]
Final design
Seven bays, one camera hood, one display.
7 independent bays
AI camera hood
Glanceable status























