When a robot deployment struggles, attention goes to the robot. The higher-leverage move is almost always changing the task so the robot already available can succeed — and that change is usually cheap.
Reduce variation at the source
The principle behind everything else in this article.
Robots handle repetition well and variation badly. Every source of variation removed makes the task materially easier, often more than upgrading the machine would.
Variation in item position. A tray with compartments instead of a loose pile. A marked spot instead of anywhere on a bench.
Variation in item condition. Ask upstream to present items consistently — same orientation, same packaging, clean. This is frequently free and it is the highest-return change available.
Variation in the environment. Consistent lighting, clear routes, furniture that stays where it is.
Variation in timing. Predictable arrival rather than bursts.
The counter-intuitive part: the department that can most cheaply solve a robot's problem is often not the one operating the robot. A change upstream in how items are delivered can eliminate a problem that would otherwise require expensive sensing.
Prepare the physical space
Cheap changes with disproportionate effect.
Define one clear route. Not the whole space — just one path the robot uses. Mark it, and keep it clear.
Remove fixed obstacles from that route. Bins, planters, standing signs. Each one is a point where the robot slows or stops.
Fix furniture positions. Where chairs are pulled into a route, arrange them so that even pulled out they do not encroach.
Control lighting where vision is involved. Dedicated lighting shielded from daylight variation. This single change resolves a large share of vision problems.
Handle thresholds and steps. A small ramp costs very little and can open an entire area.
Provide convenient charging. Out of the traffic path, near where the robot works, with a clear routine for who plugs it in.
Give it somewhere to park. A robot blocking an entrance overnight gets pushed into a storeroom.
Change the process, not just the layout
Larger changes with correspondingly larger effect.
Move the decision to a person. Where a task requires judgement at one step, have a person make that judgement and the robot execute the rest. Splitting a task this way often converts an impossible job into an easy one.
Batch similar work. Grouping like items reduces changeover and variation simultaneously.
Standardise across products. One tray type, one packaging format, one presentation method — even if slightly suboptimal for each individual product.
Add a verification step. A sensor confirming the previous step succeeded turns a compounding error into a controlled stop.
Shorten sequences. Break a long dependent chain into short independent operations with checks between them.
Change what upstream sends. The single highest-return change and the one most often not attempted because it involves another department.
Design the human side too
Half the failures in real deployments are about people rather than machines.
Define who does what. Written down and posted. Ambiguity in the first weeks becomes habit.
Define the handover point. When the robot passes to a person, how, and what information travels with it.
Make the correct behaviour the easy behaviour. If keeping an aisle clear requires effort, it will not happen. Provide somewhere for the items that currently block it.
Establish new habits deliberately. Unloading a tray promptly, keeping a route free, plugging in at the end of a shift. These take a few weeks and they need reinforcing during that period.
Give people a way to report problems. Simple enough that they actually use it — a sheet by the desk works.
Name an owner. Without one, everything above decays.
A practical sequence
How to apply all of this before buying anything.
One: watch the task for a full shift. Not a sample — a whole shift, noting every variation and every interruption.
Two: list every source of variation. Position, condition, timing, environment, product mix.
Three: cross off the ones that are cheap to remove. Usually more than expected.
Four: check whether the remaining task is now within reach. Often it is.
Five: only then evaluate equipment, against the simplified task rather than the original one.
Teams that follow this sequence buy smaller, cheaper equipment and get better results than teams that specify a robot capable of handling the task as it currently exists.
And there is a secondary benefit worth noting: most of the variation-reduction work improves the task even if no robot is ever purchased. Clearer routes, consistent presentation and defined handovers make manual work better too.
Frequently asked questions
What is the highest-return change in most deployments?
Asking upstream to present items consistently — same orientation, same packaging, clean. It is frequently free and it eliminates problems that would otherwise require expensive sensing.
Which environmental change resolves the most vision problems?
Dedicated lighting shielded from daylight variation. A vision system relying on room light behaves differently by season, time of day and whether a door is open.
How can an impossible task become an easy one?
By moving the judgement step to a person and having the robot execute the rest. Splitting a task at the point requiring judgement frequently converts something out of reach into something routine.
What is the benefit if no robot is ever bought?
Most variation-reduction work improves the task regardless — clearer routes, consistent presentation and defined handovers make manual work better and faster too.
More in Local ecosystem and Voice and language.
