Mission library · Industry & logistics

Site patrol robots that never leave coverage.

A patrol route is chosen by what needs watching — the gate, the tank farm, the dark side of the warehouse — not by the Wi-Fi survey. This mission is about removing the network from the route-planning meeting: facility Wi-Fi and cellular bonded into one tunnel, so every seam the robot crosses is a re-weight instead of a reconnection.

Who runs it

Industrial site operators, security integrators, and operators of quadruped and wheeled UGV patrol fleets.

What breaks

The route crosses Wi-Fi cells, yard dead zones and perimeter stretches where only cellular reaches — each seam stalls the patrol or drops the feed.

What Atlas contributes

Facility Wi-Fi and cellular bonded into one tunnel with a fixed address; endpoint roaming across every seam; a relay node for the dead corners.

Runs on

The robot’s existing compute and the control-room workstation — one Linux binary, one configuration file, no new radio hardware on the robot.

The mission

A chemical storage site runs two patrol robots on night rounds: gates and fence line every hour, the tank farm twice a shift, thermal sweeps of the warehouse loading faces. The control room — sometimes on site, often a monitoring centre in another city — watches the feeds, answers alarms, and occasionally takes a robot under manual control to look at something properly. The value case is simple: the robots walk the boring kilometres so the guards handle the judgement calls. The whole arrangement stands on one assumption — that the control room can see and steer the robot wherever the route goes, all night, every night.

What breaks

An industrial site is a patchwork of networks that were never designed as one. Facility Wi-Fi covers the buildings and dies in the yards; the access points hand clients over roughly, and a robot trotting between cells is the roaming client every enterprise Wi-Fi deployment handles worst. Cellular fills some of the yard and the perimeter, with its own dead spots behind the steel. A robot on a single network therefore lives a stop-start life: the video feed resets at every AP handover, the session drops in the yard, and the teleoperation session — the thing the guard needs during an incident — is least reliable exactly where incidents happen, at the fence and in the dark corners.

The classical fixes each have a bill attached. Blanketing the yards in outdoor Wi-Fi is a capital project with trenching in it. Constraining the patrol route to the coverage map inverts the priorities — the network now decides what gets watched. And per-network client tuning is a treadmill: every AP firmware update, every new cell, another seam to debug.

The architecture on this mission

Site patrol coverage-seam geometry A patrol robot crossing the dead zone between two facility Wi-Fi cells stays connected over cellular; the control room’s feed continues through the seam. the yard — where the route goes and the Wi-Fi does not WI-FI CELL A WI-FI CELL B DEAD ZONE CELLULAR THE SEAM IS A RE-WEIGHT, NOT A RECONNECT Control room watching the feeds Patrol robot
The scene: mid-seam, the Wi-Fi cell behind the robot is gone (crossed) and the cellular path carries the same tunnel on to the control room. Back inside the next cell, Wi-Fi re-scores its way into the bond and the cellular spend drops away — the feed never resets.

The robot’s node bonds everything the site offers — the facility Wi-Fi interface it already has and a cellular modem — into one encrypted tunnel with one fixed address. The control room talks to that address and nothing else; which physical network happens to be carrying at any given metre of the route is Atlas’s problem, not the video client’s. Every link is probed several times a second, so a weakening Wi-Fi cell is measured and de-weighted before the handover, and the rough AP-to-AP transitions that break ordinary clients become internal re-weights: endpoint roaming carries the session across every re-association and address change, and the control room’s feed simply continues.

The corners neither network reaches — behind the tank farm, inside the far warehouse aisle — are covered the way the mesh covers everything: with another node. A fixed relay box on a mast, or simply the second robot holding at the seam for the minute its partner needs, extends the fabric around the steel; relays cannot read the traffic they carry. Traffic classes keep the priorities straight when a seam leaves only a thin path: teleoperation and command hold a reserved floor, the video sheds quality before continuity, and bulk uploads — the night’s recorded patrol video — drain in the background without ever contending with the feed the guard is actually watching.

How the night shift unfolds

  1. Rounds begin. The robot leaves the dock on warehouse Wi-Fi. The tunnel to the control room is already up; the cellular link idles at a trickle, measured and ready. One dashboard shows both links per robot.
  2. The first seam. Between building cells, the AP handover that used to reset the feed happens inside the bond — a re-weight logged in the journal, invisible in the control room.
  3. The yard. Wi-Fi fades on schedule as the robot crosses to the tank farm; traffic is already flowing on cellular before the last probe fails. The thermal sweep streams uninterrupted from the middle of the dead zone.
  4. The dark corner. The route’s one true hole sits behind the tanks. The fixed relay node on the mast — or the second robot holding at the seam — carries the hop around the steel, and the corner gets watched instead of skipped.
  5. The incident. A gate alarm at the fence line. The guard takes manual control; teleoperation rides the reserved floor across whatever mix of Wi-Fi and cellular the fence offers, and the video coarsens rather than freezes while the robot closes in.
  6. The morning after. The journal holds a per-link, per-lap record of the whole night — every seam, every re-weight, every metre of single-path operation. Over a month it is a measured coverage map of the site: the case for one more AP, or proof none is needed.

What each mechanism contributes

  • Wi-Fi + cellular bonding — one tunnel, one fixed address per robot; the control room never chases the robot across networks. Shipped.
  • Endpoint roaming — AP handovers, re-associations and address changes absorbed inside the tunnel; sessions survive the seams. Shipped.
  • Mesh relaying — a fixed node or a second robot extends the fabric into the dead corners, with per-hop encryption. Shipped.
  • Class floors and ordered degradation — teleoperation reserved, live video elastic, bulk patrol footage last. Shipped, opt-in.
  • Continuous measurement and the journal — a per-seam record of every lap, accumulating into the site’s real coverage map. Shipped.

The honest boundary: Atlas cannot manufacture coverage in a corner that has neither Wi-Fi nor cellular nor a relay in reach — it makes the hole measurable and small, and the relay node placeable, but the site must field that node. The robots, their autonomy stack, the patrol doctrine and the response to what the cameras find remain the operator’s.

What a pilot should prove

  • Feed continuity across a full patrol lap — every Wi-Fi seam and the yard crossing — bonded versus the robot’s stock single-network setup.
  • Teleoperation held through the worst measured seam on the route, on the reserved floor, while video sheds quality rather than freezing.
  • One dead corner closed by a relay node, with the per-hop record showing it carrying.
  • A month of journals condensed into a measured coverage map the security manager can plan the next route from.

One site, one robot, one full shift pattern. The evaluation format covers the structure.

Request a briefing

Fielding a patrol robot fleet?

Bring the robot, the site’s Wi-Fi map and the one corner every route avoids — we will define what a pilot deployment should prove.