Robot-to-Robot Handoff
At a glance: Two robots working the same building rarely fail on navigation. They fail at the handoff: the point where one unit leaves a load or a task for another. That handoff is not in either vendor's scope of supply, it is not in either contract, and it is the line item that turns a straightforward deployment into a six-month integration project. This guide prices the three physical handoff modes, specifies the message layer that makes them reliable, and gives the acceptance test that proves the handoff works before the fleet is accepted.

Why Multi-Vendor Fleets Fail at the Handoff
A robot-to-robot handoff is any point where responsibility for a payload, a task or a zone passes from one unit to another, and in most deployments nobody is contractually responsible for it. A delivery robot vendor warrants its robot; a cleaning robot vendor warrants its robot; a lift controller vendor warrants its controller. The moment a parcel moves from the delivery unit to a locker robot, or a cleaned zone is handed to a floor-scrubber for a second pass, the integration between them belongs to the operator, who did not write either specification.
The failure is commercially predictable. Ask three questions in the tender: who owns the handoff logic, what message format the two systems exchange, and which party pays if a handoff fails halfway through. If all three answers are "the integrator" or "to be agreed", the project has an unpriced scope that will surface at commissioning.
This article treats the handoff as an engineering and a commercial object: three physical modes with different cost structures, one message contract that covers all three, and an acceptance test that can be run in a day rather than discovered over a quarter.

The Three Handoff Modes and What Each Costs
Pick the physical mode first, because the mode determines the cost and the failure modes.
| Mode | How it works | Typical incremental cost | Dominant failure |
|---|---|---|---|
| Fixed transfer station | Both robots dock at a bench, shelf or conveyor and the payload is transferred by gravity, a powered roller or the receiving unit's own arm | 2,500 to 9,000 for the station, plus commissioning | Positioning tolerance: the two dock poses differ by more than the transfer gap |
| Direct unit-to-unit transfer | One unit presents the payload and the other collects it without a station in between | Near zero hardware, highest software cost | Simultaneous occupancy of a single pose, and no fallback if either unit is late |
| Zone handoff, no payload | Responsibility for an area passes between units, for example a corridor cleaned by a large scrubber and finished at the edges by a compact unit | Lowest: a state record, no physical interface | Double coverage or a gap at the overlap, invisible without completion telemetry |
The fixed station is the cheapest to make reliable and the most expensive to install. Direct transfer is the reverse and is only sensible between units of the same fleet and the same generation. Zone handoff costs almost nothing physically and fails silently, which is the worst combination because the failure appears as a quality complaint six weeks later rather than as an integration error on the day.
Where the two units are the same model from the same generation, the design problem collapses to identity and handshaking, and the platform's own fleet layer can carry it. Where they are different models or different vendors, the problem is genuinely an integration and should be budgeted as one. The multi-vendor fleet interoperability guide covers the vendor-side API and data-ownership terms that make the second case tractable.
One arithmetic rule governs all three: the handoff cycle time must be shorter than the dispatch interval, or the queue in front of the handoff point becomes the bottleneck for the whole fleet. A handoff that takes 140 seconds against a 100-second dispatch interval converts two robots into one plus a growing backlog.
The Handoff Contract: Six Messages
Whatever the physical mode, the two systems exchange the same six messages. Specifying them explicitly is what allows a handoff to be tested rather than hoped for.
| Message | Sender | Content | Required response |
|---|---|---|---|
| offer | Offering unit | Task or payload id, weight, dimensions, target zone, deadline | Acknowledgement with accept or refuse inside a stated window |
| accept | Receiving unit | Acceptance, promised start time, receiving pose | Offerer commits the payload and stops accepting new work |
| arrive | Offering unit | Position at the handoff point, payload state, integrity check result | Receiver begins its approach sequence |
| transfer-start | Either | Timestamp, both unit ids, mode, sensor confirmation | Both units lock the pose and inhibit unrelated motion |
| transfer-complete | Receiving unit | Payload id confirmed, condition, timestamp | Offering unit releases and clears its record |
| abort | Either | Reason code, last known good state, safe position | Both units enter a defined degraded state and escalate |
Three properties make this set work in practice. Every message carries the payload identifier, so a failure is attributable to a physical object rather than to a robot. Every message has a timeout with a defined action, so a silent unit is handled by the protocol rather than by an operator with a phone. And the abort message is mandatory rather than exceptional: a system without a specified abort path will invent one at runtime, usually by stopping in place.

Zone Handoff Without a Payload: The Silent Failure
The most common multi-vendor deployment is not a parcel transfer; it is two cleaning units dividing a building. A large scrubber covers open floor and a compact unit handles edges, lifts and toilets. Nothing is physically handed over, which is why this mode is assumed to need no integration at all. It needs three records.
- Coverage record. Each unit reports the area it actually completed, not the area it was routed to. A route completion flag reports intent; a coverage record reports work.
- Overlap rule. The shared boundary needs a defined overlap, typically one metre or one pass width, and one unit must be designated the owner of the overlap. Without a rule, either both clean it or neither does.
- Exception record. Obstacles that blocked coverage, with position and reason, so the receiving unit can be dispatched to them rather than re-running the whole zone.
These three records cost a data schema and a merge step. Their absence is why operators report that a building is "clean but not properly clean" after a multi-unit deployment, a claim that cannot be resolved because neither unit's report is wrong on its own terms. The verification view of the same problem, including how a completion record is audited independently of the unit that produced it, is set out in the cleaning robot verification and audit trail guide.
Time, Queueing and the Cost of a Slow Handoff
Handoff economics are queue economics, and the arithmetic is short enough to do before purchase.
| Parameter | Worked example, 14 drops per hour |
|---|---|
| Dispatch interval | 3600 / 14 = 257 s |
| Handoff cycle time | 95 s physical, plus 25 s message and confirmation = 120 s |
| Handoff utilisation | 120 / 257 = 47 percent |
| Queue at 47 percent utilisation | Average 0.4 units waiting; 90th percentile about 1.9 units |
| Same system at 20 drops per hour | Interval 180 s, utilisation 67 percent, 90th percentile queue about 4.1 units |
The queue grows non-linearly as utilisation approaches one. A handoff that is comfortably fast at pilot volume becomes the constraint at production volume, and the operator experiences it as "the fleet needs more robots" when the real constraint is a single station. The payload capacity and fleet sizing guide carries the same arithmetic from the capacity side, and the two should be done together: unit count and handoff throughput have to be sized as one system.
The cost of a slow handoff is therefore not the idle time of one robot. It is the extra unit the operator buys to compensate, which is a capital line item that appears in year one and a service contract that persists for the life of the fleet.

Contract Structure: Who Owns the Seam
Three structures are available and each has a clear consequence.
| Structure | How it works | Best fit |
|---|---|---|
| Single prime integrator | One party supplies both fleets and the handoff layer, and warrants the end-to-end task | Greenfield deployments where neither fleet is already owned |
| Interface specification plus test | Two vendors each warrant their own unit and both warrant conformance to a published interface test | Existing fleet plus a new vendor, where the interface can be tested independently |
| Operator-owned middleware | The operator builds and owns the handoff layer, and vendors expose APIs | Large estates with an in-house integration team and several vendors |
The second structure is the most common and the weakest in practice, because conformance to an interface test does not prove that two independently conformant systems interoperate. The remedy is to test the pair, not each side, and to put the pair test in both contracts. Where a buyer is procuring two fleets at once, the cheapest clause to write is that the two vendors must jointly demonstrate a handoff during acceptance, with the failure cost borne by the party whose unit failed to respond within the timeout.
Support and escalation terms matter here as much as the technical interface. When a handoff fails at 2 a.m., the question is who is paged and who is contractually obliged to diagnose first. The uptime SLA contract guide covers response-time credits and the escalation path, and those terms should be applied to the handoff layer explicitly rather than inherited from the unit warranty.
Acceptance Test: Proving the Handoff
A handoff acceptance test does not need a week. It needs the pair of units, a stopwatch and a fault injector, and it should be run against volume, not against a single happy path.
- Steady-state run. One hundred consecutive handoffs at production dispatch interval, with the count of successful completions, average cycle time and 95th percentile cycle time recorded. A fleet that averages 120 seconds but has a 400-second 95th percentile will queue at peak.
- Late-arrival injection. Delay one unit by 90 seconds in ten of the hundred runs. The protocol must refuse, reassign or abort cleanly; the failure mode to watch for is both units waiting indefinitely for each other.
- Silent-unit injection. Disconnect the communication of one unit mid-transfer. The other must reach a defined safe state within the stated timeout and raise a single attributable alarm, not a storm of retries.
- Restart reconciliation. Power-cycle both units after a transfer-complete but before the record was cleared. On restart the system must resolve the payload to one owner, not zero or two.
- Volume stress. Run at 130 percent of the planned dispatch rate for fifteen minutes and record how the system degrades: does it queue, refuse new work, or accept work it cannot complete?
The battery swap station design guide covers how pack identity is preserved across a swap, which is the physical event that most often triggers the state loss this test targets.
The restart reconciliation test is the one that finds the defects nobody planned for. Handoffs look robust in a demonstration and reveal their state-handling problems only after an unplanned reboot, which in service happens whenever a pack is swapped or a unit faults.

Pricing the Integration as a Line Item
Buyers who price the handoff explicitly get fewer surprises. A realistic budget for a two-vendor handoff breaks into five parts: interface documentation and adaptation on each side, the middleware or orchestration layer that routes the six messages, the pair acceptance test, contingency for a protocol change mid-project, and the ongoing support term. The middleware and the pair test are the two lines that are almost always omitted, and they are the two that determine whether commissioning finishes on schedule.
The alternative to pricing it is discovering it. An unpriced handoff appears at commissioning as an integration overrun, gets absorbed by whoever is on site, and usually ends with a manual workaround such as staff carrying the payload across the seam. That workaround is invisible in the deployment report and permanent in the operating cost.
How AOMAN Approaches Fleet Interoperability
AOMAN supplies the D1 delivery, C1 and C2 Pro cleaning, and G1 reception platforms with a documented integration layer rather than a closed appliance, and publishes the message and data contracts a buyer needs to build or procure a handoff layer. Where an AOMAN platform is paired with a third-party unit, AOMAN provides the interface documentation and the acceptance-test procedure for its side of the seam, so that the pair test is a scheduled exercise rather than an open-ended integration. Details of the platforms are on the products page, and the fleet-level architecture is described on the technology page; the multi-vendor programme view is covered in the fleet interoperability guide.
