Uncrewed & autonomous systems

The platform is autonomous. The link shouldn’t be the weak point.

For UAV, UGV and mobile-robot operators and OEMs: command, telemetry and video that keep moving as the platform changes range, coverage and direct reachability — on the companion computer it already carries.

~15 g compute floorNo airframe redesign

The stakes

The stakes
For platforms in motion, the killer is rarely a broken radio — it’s geometry.
01

The video dies at the range boundary

The aircraft flies on, but the short-range link it favoured has stretched past its reach — and the handover happened after the freeze, not before it.

02

The AGV fleet stalls in a radio shadow

Between two buildings, one access point’s coverage ends and the next hasn’t picked up — and every vehicle in the seam stops.

03

A full day’s data leaves coverage with the vehicle

The survey platform works beyond the network by design. The data it carries shouldn’t need an operator and a USB stick to come home.

One connection over everything

The links the platform carries, as one system.

Nexus Atlas bonds the dedicated radio, cellular modems, Wi-Fi and mesh links an uncrewed platform already has into one encrypted connection — transparent to the autopilot, the video pipeline and the mission software.

dedicated C2 radiocellular ×N802.11s meshWi-Fisatellite (where carried)
Three pillars

Built around motion.

Each pillar is a demonstrable behavior on the actual platform.

01 · PREDICT

Hand over before it drops

Shared position and velocity plus radio signal trends move traffic to the longer-reaching link before the short one fails — the handover happens while both still carry. Built, opt-in.

02 · FIT

Fits the airframe

A ~15-gram compute floor, one binary, and adapters that pull telemetry straight from the radio. No enclosure, no mounting, no added mass beyond the board the platform likely already carries.

03 · EXTEND

Every vehicle extends the network

Each platform is a relay: the fleet is the infrastructure, self-healing the moment a change is detected — ≈1.5–1.75 s at default settings for a lost vehicle.

~15 gSingle-board-computer floor for a full Atlas node.
0.1–0.75 sTypically configured reweighting cadence as conditions change.
8 hopsRelay depth through participating fleet nodes.
6 typesDissimilar link types bonded on one node in the field demo.
Mixed-route UAV

Dual radio plus cellular across a changing route.

In covered areas cellular dominates; in open country the dedicated radio carries; the scheduler’s continuous scoring shifts the mix without a switchover event. Where the aircraft is reached both directly and through a relay, its commands take whichever path has room instead of queuing behind the video. The triplex pattern in the playbook

Where both ends run Atlas, the telemetry radio carries Atlas directly — no PPP, no IP plan per radio pair, about 29 bytes less on every voice packet — and writes are paced on the radio’s own buffer, so a command or a voice frame overtakes video in Atlas’s queue instead of waiting behind it in the radio’s: voice stays usable beside video on a 64 kbit/s radio. Atlas directly on serial radios

Warehouse AGVs

Across the dead zones, through each other.

Vehicles bond Wi-Fi and mesh and relay for each other — a shadowed vehicle reaches the controller through a neighbour instead of stopping.

Field machinery

The data comes home by itself.

Store-carry-forward holds the day’s data through coverage gaps and drains it — at full bonded throughput — the moment the vehicle re-enters range or passes another node.

The same holds for video. The live feed bends to the link and drops late frames; the vehicle keeps every packet, the ground asks for exactly what it missed, and the missing pieces come home on the lowest-priority lane, behind commands and live video, during the run or after it. Only the gaps cross, never the whole recording again. How the gaps are filled

Ridge crossing

The hill is on the map. The handover is early.

A UAV runs toward rising terrain carrying a fat 5 GHz link and a thin long-range radio. Projected geometry and the elevation data loaded on the node raise the short link’s risk while it still carries, and if the aircraft drops behind the crest the route moves through a relay that still sees it within about half a second, because the aircraft shares each new position as it flies; video thins toward the long link, and by the time the ridge cuts the beam, nothing is left on it to lose. With two cameras aboard, the primary feed outranks the thermal one inside the video lane, so when the long link cannot carry both, the thermal camera gives way first. The mechanics, with a live simulation The ridge, on the engineering site

Customer networks

Your ground station, their drones.

A customer’s fleet already runs its own network. Your control station joins it as a member under the customer’s rules and bridges it to your own: your operators reach the aircraft the customer chose to share, at addresses of your network, with nothing changed on board. The customer sees none of your topology and can revoke your station like any member. Opt-in. How the seam works

A customer who must only watch the mission gets a link, not a seat in your network: the feed gateway — a separate package beside Nexus Atlas — sends one feed outward from a read-only member to a feed server; the customer opens it in a browser with a token for that feed, until it expires or is cut with one click, and the player draws the viewer’s name across the picture. Your fleet opens no port, and nothing travels inward. How the gateway works

Fleet software

Preinstalled on the aircraft, operated from a web ground station.

The aircraft ships with Atlas in its companion image and two or three cellular modems. It dials out over every modem to two hubs in front of the ground station’s backend; commands ride the two best modems at once, video is paced behind them, and a removed aircraft is refused everywhere within seconds. A new aircraft joins from a pairing code the ground station shows, or by a certificate signed at the factory. When the hub that admits it cannot sign the fleet’s list, it hands the new aircraft to one that can, and the rest of the fleet knows it from the signed list about a second later. Each customer is a network of its own. Shipped. The pattern, for platform makers The reference architecture — nexusatlas.io

Fleet updates

A push never strands an aircraft.

A fleet update that half the fleet took is the one you find out about in the air. Every aircraft record carries an update class: a release is complete only when each required aircraft has confirmed it; one silent or failed at the deadline withdraws the change on its own — the previous settings go back out, signed — and the abort names the aircraft; best-effort aircraft take it when next seen and never hold anyone up. Keys and the shared secret rotate on the same kind of signed schedule, in windows an aircraft out of contact can still catch: one back after a week comes in on its old key, receives the new one over its first connection and switches by itself. The release gate Keys without a flag day

Pure software. Pilot on the companion computer you already fly — no procurement. Vendor-neutral by design: the radios stay yours, the carriers stay yours, and Atlas bonds whatever they provide. Predictive routing in the catalogue.
Request a briefing

See it survive a link kill, live.

Bring the airframe’s bearer mix and the mission profile — we’ll define what a demonstration on your platform should prove.