Discovered on arrival
Units are picked up from the live streams they already publish. Eighteen models are described out of the box, each using the topic names its own manufacturer actually chose.
Fleet operations
Mixed fleet
Mixed fleet
orchestration
One control plane for every robot you operate, whoever built it. You submit the work rather than the command, and the platform decides which unit runs it, drives it to completion, and records the decision it made.
Every vendor ships a dashboard.
Nobody ships a control plane.
An operator running four robots from three suppliers has four screens, three logins, no shared view of the building, and no single answer to the question that always gets asked eventually: what actually happened at 09:49. Fleets rarely stall because the robots cannot drive. They stall because nothing coordinates them and nothing can account for them afterwards.
Watching robots is table stakes. The part worth paying for is the system that decides which unit runs what, in what order, with what charge remaining, and can account for every one of those decisions afterwards.
The difference between watching a fleet and running one.
Dispatch one task, three candidates
Submit the work
Not the command
Best-fit assignment weighs availability, proximity, and whether a robot has the charge to finish the job as well as to start it. ROS 2 units and VDA 5050 warehouse AMRs join the same fleet and are dispatched, measured and audited the same way.
What the platform does
A complete operations layer above your robots: discovery, dispatch, telemetry, language and audit, delivered as one system with one API.
| Automatic discovery | Robots are detected from the streams they already publish, so there is no per-unit configuration file to write. Eighteen models are described out of the box, each with the topic names its own manufacturer chose. |
| Cross-vendor connectivity | ROS 2 units and VDA 5050 warehouse AMRs join the same fleet over their native protocols and are treated identically for dispatch, metrics and audit. |
| Best-fit dispatch | Submit an outcome and the platform assigns it, weighing availability, proximity, and whether a unit has the charge to finish the job as well as to start it. |
| Operational telemetry | Uptime, distance, utilisation, battery drain, navigation success and any custom measure you define, held as durable history that is queryable and exportable. |
| Natural-language operations | Operators state intent in plain language against the named places in your building. With no model available at all the same commands still work, through a deterministic planner. |
| Append-only record | Every registration, assignment, command, reroute and completion, written as it happens and immutable afterwards. |
What the record looks like
The second line is the one that matters: the platform records not only what it chose but what it declined to choose, and why.
Fleet record excerpt, append-only
01
Task accepted ward 4 → path lab
Routine priority, requested from the ward screen
QUEUED09:41:02
02
Assigned to unit 2
Unit 1 was 12 m closer and was passed over.
34% charge would not have completed the return leg. Unit 2: idle, 91%, 34 m out.
ASSIGNED09:41:02
03
Leg 1 complete ward 4
Payload taken on, compartment closed
PICKED UP09:47:55
Re-planning · in transit
East corridor closed by facilities at 09:49. The same closure that moved the
visitor journeys; this robot was on it.
04
Leg 2 complete path lab
Delivered and handed over at receiving
DONE+70 m · 09:58:31
A fleet nobody can account for is a fleet that gets switched off the first time something goes missing. The record is written as the work happens, and it cannot be rewritten later.
Two standards, one fleet
ROS 2
VDA 5050
Warehouse AMRs, same fleet
The industry standard for automated mobile robots joins over MQTT and is dispatched and audited the same way, inside the same console and the same reporting.
Anything else
One connector per vendor
Everything normalises into a single internal model of what a robot is, so a vendor with a proprietary API costs one adapter, and each integration after it gets cheaper.
Safety stays where it belongs
The boundary is deliberate
- The platform decides what work goes where. It will never sit between a robot and its own certified safety stack. That system stops the machine, and nothing we ship is permitted to override it.
- We orchestrate, and we stay out of the safety path. That is what lets the platform drop into a fleet without re-opening anyone’s safety case.
How many screens is your fleet on?
If the answer is more than one, or if somebody has asked you for a record of what the robots did last month and you had to assemble it by hand, we should talk.