Facilities AI agent · Cleaning robotics
Cleaning Robotics Coordination AI Agent
Confine a cleaning run to the zone confirmed clear on a live occupancy read, carry signage and drying time inside the run itself, and hand the floor back under the operations supervisor.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Zone-bound
- People first
- After launch
- Agent Care
What this agent does
Schedules the run, never overrides the zone
In
- 01
A run is scheduled, and the zone it covers must read clear on a live occupancy check first.
- 02
A zone is closed by somebody on the floor, and it stays closed until that same person reopens it.
Reason
- 03
A machine loses its position, and it stops where it stands rather than working on dead reckoning.
- 04
A wet floor goes down, and the signage, the drying time and the reopening belong to that run.
- 05
An interlock trips, and the fleet is halted rather than routed around the fault that tripped it.
Decide
- 06
A shift hands over, and every zone a supervisor shut carries across the handover still shut.
- 07
A route crosses a corridor in use, and right of way stays with the people already walking it.
Out
- 08
A zone is called cleaned, and the run record has to carry the coverage that claim rests on.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent proposes runs and tracks the zones. The operations supervisor owns the zoning and interlock rules, and releases each run inside them.
Example workflow
One run, proposal to release
- 1Run conditions receivedOccupancy reads, zone states, machine telemetry or interlock signals
- 2Run context assembledThe zone, the machine put on it, the rules it answers to and the hour the floor comes free
- 3Run plan draftedThe zone, its occupancy reads, the drying window and completeness
- 4Controls appliedOccupancy checks, interlock checks, localisation checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is released at any of them — the agent is planning, and the operations lane opens at the completeness gate.
Goes to the operations supervisor to release.
Adds a cleaning lead read first.
Operations review
The plan is held with its zone, its occupancy reads and the machines it would commit.
Release · Append conditions · Send to operations review
Released — by the operations supervisor
- 6Fleet and zone records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedZone adherence, coverage evidence, reviewer corrections and what the read found
Corrections
Each operations correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Overriding an interlock the supervisor set.
- Reopening a zone somebody on the floor closed.
- Working a machine that cannot place itself.
- Deciding when a wet floor is walkable again.
Automation boundary
Agent acts unaided
- ✓Hold every run to a zone that reads clear on the latest.
- ✓Carry the signage, the drying time and the handover inside the run.
- ✓Halt the fleet the moment a machine can no longer.
- ✓Show the coverage a run achieved against the zone it was given to cover.
Nothing is released or reopened except by a named person, inside the agreed boundaries.
- Judging whether a floor is dry enough to reopen.
- Telling the site that a wing was cleaned.
- Setting the zoning a machine may work inside.
- Changes to interlocks, zone maps or route rules.
Example output
One run, annotated
Our intralogistics robot fleet agent sequences material movement inside plant zones and speed limits; this is cleaning plant sharing a floor with people, where an occupied zone and a wet floor bind before throughput does.
Run record · single zone
Illustrative example- Run
- Scrubber run, level two east
- Recorded as
- Covered in part, zone returned late
- Zone
- Level two east
- Evidence of record
- Occupancy read, 3 August 2026
- Confidence
- Held unreleased
- Held for
- The operations supervisor, by name
As received
Taken from the machine telemetry and the occupancy feed — it reaches as far as those sources do.
What the record holds
- Machine telemetry
- Occupancy read log
- Wet-floor sign log
Why no reopening here
When a wet floor may reopen is a supervisor judgement, not a model output.
What the score decides
Below the configured threshold a plan picks up a cleaning-lead read before release.
Action
ReleaseAppend conditionsSend to operations reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every run
From the floor it works on
Where the evidence is used
A cleaning schedule is not a cleaning record: this page holds what each run actually covered, and it claims nothing about how clean the floor was left behind.
The floor is shared
Right of way sits with the people on the floor, so an occupied zone is a closed zone and no run is scheduled into one.
What was checked, and not found
No cleaning-machine safety standard was bought or read, no vendor claim about obstacle detection or localisation accuracy was accepted, no productivity benchmark from a trade survey was carried onto this page, and what the agent may hold about a person a machine passes is governed by your own configured policy.
Build an evidence trail
The run, the zone it covered and the supervisor who released it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Fleet and machine sources
- Robot fleet manager
- Machine telemetry and faults
Occupancy and access
- Badge readers
- sensors
- Zone occupancy reads
Building and safety systems
- BMS
- interlock panels
- Zone states and closures
Agent
Cleaning robotics coordination
- Reads the zones
- Plans the runs
- Holds for the supervisor
Work order and audit systems
- CAFM
- inspection sheets
- Run records and coverage maps
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six traps between the model and the supervisor
Six traps set along the route, the last hardest to pass. Whatever survives is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a run may stand
- L3
- Puts the release in a person's hands
- L4 – L5
- Keep the run and the zone behind it
- L6
- Returns the fleet to manual when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the run plan that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the record a supervisor reads
- E1Final-output evaluationDid the record show the coverage the run actually achieved?
- E2Step-level evaluationDid the agent read the right zone, the right machine and the live occupancy?
- E3Tool evaluationDid it read and write the correct zone record and the correct run?
- E4Confidence calibrationDo ageing occupancy reads actually attract more operations corrections?
- E5Slice evaluationHow does performance change across specific fleet units?
- E6Business outcomeHow many plans needed a correction before the supervisor released?
Floor — the run record the site answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each caught at the stage where it first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- KT-03
Stale occupancy read
The zone state read is not the one now on the floor.
Stage gathers
The zones, the runs, the units and the reads
02 · Reasoning
2 modes
- KT-04
Run called done, not covered
A zone is reported clean without its coverage.
- KT-06
Closed zone read as open
A zone a person shut is worked as available.
Stage proposes
The runs, their zones and coverage
03 · Tool / write
2 modes
- KT-02
Thin plan passed forward
A plan moves on without the operations read.
- KT-05
Filed against the wrong unit
A run is recorded against the wrong machine.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- KT-01
Released, coverage unrecorded
The record carries a release but not what the run covered.
Stage returns
The run record a supervisor reads and a site keeps
05 · Change / Version
1 mode
- KT-07
Silent zoning regression
A map change moves the zone, not the run rules.
Stage tracks
Model, prompt, zoning rules and run fields
- Sev-1 · a run released into an occupied zone
- Sev-2 · a closed zone entered by the fleet
- Sev-3 · source degrades, run held unreleased
Affected slices
Scrubber-driers absorb the corrections
A unit-level zone-adherence figure can read clean while scrubber-drier units carry most of the rework. Nestack reports the correction rate by fleet unit, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Scrubber-drier units | 6.6% | 3.6× | Review | |
| Autonomous vacuum units | 4.7% | 2.6× | Review | |
| Escort-mode sweeper units | 2.9% | 1.6× | Watch | |
| Static disinfection units | 1.6% | 0.9× | Normal |
Bar: correction-rate lift vs. static disinfection unit baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an occupied zone costs
A cycle ends when the wet-floor overlap is a standing case. That suite is what the next run scheduled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on scrubber-drier units.
- 02
Diagnose
The machine that washed a floor a shift was still standing on is traced back until a single cause remains.
- 03
Improve
Changes leave numbered, and the runs that forced them travel attached.
- 04
Verify
One run case still failing is enough to stop the whole release.
- 05
Learn
The case is kept, and the zoning rules are rewritten alongside it.
Learn → Detect
The return edge. The next run is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, run planning, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Zone-map discovery and automation-boundary work.Week 1
- 02Fleet, occupancy and zone-map sources.Week 1, Week 2
- 03Run-to-zone and coverage-evidence rule mapping.Week 1, Week 2
- 04Run and telemetry ingestion.Week 2
- 05Zone, machine and window binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Supervisor release workflow.Week 3, Week 4
- 08Fleet-management system integration.Week 5
- 09Zoning and interlock cases.Week 4, Week 5
- 10Guardrails and release controls.Week 4
- 11Run-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne site, one quarter | ProductionProduction cleaning workflow | AdvancedMultiple sites / fleets |
|---|---|---|---|
| Introduced at Pilot | |||
| Run planning to your zone map | ✓ | ✓ | ✓ |
| Operations supervisor release | ✓ | ✓ | ✓ |
| Cleanable-area baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by fleet unit | — | ✓ | ✓ |
| Release workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Fleet-telemetry integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-site fleets | — | — | ✓ |
| Cross-shift run packs | — | — | ✓ |
| Large machine fleets | — | — | ✓ |
| Multi-shift interlock controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, fleet size, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your sites and the floors each machine works
→Zone mapping and run capture
Week 102Representative run, telemetry and occupancy records
→Run binding, zoning logic and the coverage baseline
Week 203Your zoning and interlock rules
→Zoning mapping, interlock binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Fleet, occupancy and zone-map source assessment, then integration setup
Week 205Runs you would not want replayed
→Zoning cases and failure-mode testing
Week 406What no cleaning run may prove
→Coverage scoring, review routing, guardrails and release controls
Week 307A named operations supervisor to release
→Release workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The bands below are weeks of genuine effort rather than layout, so one of them has to carry two.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Cleaning workflow discovery, zone mapping and the automation boundary
W2Source integration and the cleanable-area baseline
W3Run planning, zoning logic and release controls
W4Evaluation suite, zoning cases and failure-mode testing
W5Fleet integration, pilot runs and targeted corrections
W6One cleaning quarter run under the operations supervisor, then Agent Care handover
- Reading the band
- Nothing here is stretched to fill a grid; the fifth band holds two pieces of work because it must.
- At the end of W6
- Once the run record validates, Agent Care assumes the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Facilities AI agent
Build a coordination agent around the run that began while the floor was still occupied.
Show us one run and the zone it covered. Not a schedule saying the floor was cleaned. A record of where the machine went, when the signage went up, how long the floor stayed wet and who handed the zone back to the people who walk it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your zone map
- Zone-bound
- People first
- After launch
- Agent Care available
Nestack Agents · Cleaning robotics
AGT-AF-11 · Agent Care available after launch