The autonomy stack gets the headlines — the perception, the planning, the steering. But a self-driving mining truck, port shuttle or agricultural robot is still a machine that relies on an industrial cellular router to deliver its network: for remote takeover, over-the-air updates, fleet telemetry, high-definition map and RTK delivery, and the security that keeps three hundred of them from becoming three hundred entry points. This is a field guide to that connectivity layer, treated as the distinct engineering discipline it is — one that must be designed to fail safe, never fail loud, because the safety-critical autonomy stack must keep working the instant the link drops.
Written by E-Lins Engineering Team

01 · The Reframe: Connectivity Is Not the Autonomy Stack
Start with the thing most vendor brochures blur together, because getting it wrong is how autonomous-vehicle network projects go over budget and into safety review. A self-driving vehicle contains two separate systems that happen to share sheet metal: the autonomy stack — sensors, perception, planning and the actuation that actually steers, brakes and throttles — and the connectivity layer that carries data to and from the vehicle when it is not safely parked. They are different engineering disciplines, with different failure modes, and they must be designed as separate things.
| The core thing to take from this article, if you read nothing else: an industrial router does not run the autonomy. It does not perceive the road, plan the path or command the actuators. Its job is to be the reliable, rugged, secure backhaul that the rest of the operation — the remote operator, the update server, the fleet manager, the map — uses to reach the vehicle. Treat it as a utility, and design it so the autonomy never depends on it to stay safe. |
That separation is not a footnote. It is the single design decision that lets the rest of this article make sense. When the cellular link drops — and on a moving vehicle it will drop, under a gantry crane, in a pit, behind a grain silo — the vehicle must continue to behave safely on its own. The connectivity layer is there to make the vehicle better managed, better updated and remotely supervised, not to be the thing that keeps it from driving into a wall. Hold that line and every product choice below follows.
The E-Lins in-vehicle portfolio (the H820Q, H820, H750 and H720 families explicitly list In-Vehicle among their qualified applications) is built for exactly this role: a hardened cellular router that sits in the vehicle’s electrical environment, bridges the onboard buses to a cellular uplink, and keeps a fleet reachable without pretending to be the brain. What follows is a practical tour of what that actually demands.

02 · Why an Autonomous Vehicle Still Needs a Network
If the autonomy is local, why bother with a link at all? Because an autonomous vehicle that is never connected is an autonomous vehicle you cannot supervise, update, locate or audit — and at fleet scale that is an operational and safety liability in its own right. Five jobs genuinely require a network, and none of them are the control loop.
- Teleoperation and remote takeover. Most deployments keep a human on the loop for edge cases — a confused perception, a blocked path, a site that changed since the last map. That needs a low-latency video and command uplink that survives handover between carriers as the vehicle moves.
- Over-the-air software and map updates. Autonomy software and HD maps are updated constantly. Pushing a new build to three hundred vehicles one USB stick at a time is not a maintenance plan; a managed OTA path is.
- Fleet telemetry and health. Battery state, thermal margins, actuation faults, drive hours — the data that tells you which unit needs service before it fails. This is the difference between predictive maintenance and a truck stopped in a pit.
- HD map and RTK correction delivery. Precise positioning and current map tiles are often distributed from a fleet server or a correction source rather than stored and trusted forever on the vehicle. The link is what keeps them fresh.
- Remote supervision and audit. Knowing where every unit is, what it did and whether its last autonomy disengagement was handled safely — the visibility a safety authority asks for, and a connectivity problem, not an autonomy one.
| The five reasons, in one line each |
| Takeover → Updates → Telemetry → Maps/RTK → Supervision. Notice the pattern: every one of these reaches the vehicle from outside. None of them is the thing keeping it on the path. |

03 · The Fail-Safe Principle: The Link Is Never on the Safety Path
This is the discipline that separates a vehicle network from an office network. In a building, if the WAN drops, the worst case is “nobody can reach the printer.” On a moving autonomous vehicle, a dropped link must degrade to a safe state — a controlled stop, a pull to the verge, a handover to local autonomy — and never to a loss of control. The connectivity layer has to be architected so that its absence is a planned-for condition, not an emergency.
| Honest boundary, stated plainly: E-Lins industrial routers are not part of the autonomous-driving safety system. They do not command steering, braking or throttle, and a deployment that routes safety-critical control through the cellular link is mis-designed regardless of which router is used. The router’s mandate is to be present and reliable when the link is available, and to fail out of the way cleanly when it is not. |
Concretely, that means three habits. First, the autonomy stack must hold its own safety envelope locally and independently — the link is an input to operations, never an input to the brake. Second, any remote-takeover path must be a requested handover with a local timeout: if the command stream stops, the vehicle returns to local control or a safe stop on its own, it does not freeze waiting for a packet that is not coming. Third, the connectivity must itself be redundant (we cover that in Section 05) precisely so that “the link is down” is rare — but the design still assumes it happens.
| “The first autonomous fleet I audited had a beautiful teleop console and a terrifying assumption underneath it: that the link would always be there. We spent a week proving the vehicle did the right thing with the SIM pulled. It didn’t. Fixing the fail-safe behaviour took a day; finding the blind spot took the week.” |
| E-Lins field engineer, mining deployment |
This principle also tells you what not to do. Do not put the remote operator in the path of last resort. Do not let a mapping or RTK outage stop a vehicle that has enough local state to keep moving safely. And do not treat the connectivity as a safety feature in your risk file — it is an operations feature. The safety feature is the local autonomy that keeps working without it.

04 · Vehicle-Grade Power and Protection
A vehicle is a hostile electrical environment for sensitive electronics, and it is the part of the deployment most often under-specified. The nominal “12 V” or “24 V” label hides cranking dips, load-dump spikes, reverse polarity from a crossed jump lead, and the kind of transients that only appear because a three-hundred-amp alternator and a radio share a chassis. The router you put in a vehicle has to survive that, not just tolerate a bench supply.
| What the E-Lins in-vehicle platforms actually specify |
| —Ignition sensing (automatic ON and time-delay OFF) — the router powers up with the ignition and shuts down on a configurable delay after it, so it does not silently drain a parked vehicle’s battery. Listed on the H820Q, H820 and H750. —5–40 V DC input range (5–60 V DC option), with Dual or Tri Power Inputs / Power Failover so a single feed loss does not take the link down. Vehicle installations require an inline fuse. —Reverse polarity and transient voltage protection per ISO 7637-2 — the automotive electrical-transient standard, covering load-dump and other real-world spikes. —Operating range −35°C to +75°C (storage to −40°C), a metal case, and an idle draw around 100 mA @ 12 VDC — low enough that the time-delay-off behaviour is the practical battery safeguard. |
The ignition sensing deserves a paragraph on its own, because it is the feature people forget and then regret. An autonomy computer may need to be awake during a handover; a connectivity router mostly needs to be asleep when the vehicle is parked for the weekend. Automatic ON with the ignition and a delay-off gives you both: the link is live whenever the vehicle is working, and it is not the thing that greets you Monday morning with a flat battery. For permanently-installed units (a depot gateway, a charging-pad controller) you simply power it from a always-on feed instead.
| Honest boundary on hazardous environments: E-Lins industrial routers are not explosion-proof and carry no ATEX/IECEx certification. For autonomous vehicles operating where flammable gas or dust is present — certain mining faces, petrochemical yards — the router belongs in the non-hazardous compartment (the cab, the control bay, the non-classified enclosure), with any in-zone sensing handled by certified equipment. We will say plainly when a site is outside what this hardware is rated for. |

05 · Always-On: Multi-Carrier, Modem Watchdog, Power Failover
“Always-on” is the promise teleoperation makes and the network has to keep. A moving vehicle crosses carrier boundaries, drives under coverage shadows and meets metal structures that eat signal. Single-SIM connectivity is not a plan; redundancy is. The E-Lins in-vehicle routers address this on three independent axes.
- Dual SIM, multi-carrier failover. Two SIM slots let the router hold profiles for two different mobile networks and switch when the serving carrier degrades — the single highest-leverage move for a vehicle that cannot stop to find signal.
- Advanced Modem Failure Check. The router watches the modem itself, not just the bytes, and recovers a wedged module without a site visit. A frozen modem is the most common “it just stopped talking” cause in the field, and it is exactly the one you cannot see from the NMS if the router is dumb about it.
- Power failover. Dual or triple power inputs mean a single feed loss — a corroded terminal, a blown fuse on one leg — does not drop the link. Tie the router to two independent feeds where the vehicle layout allows it.
- Link bonding (capacity, where available). Where two carriers are both strong, combining them raises usable throughput for teleop video. Treat this as a bandwidth enhancer, not the redundancy story — the failover story is the dual SIM.
| Design target, not marketing line: the goal is not “never loses a packet” — that is not physics on a moving vehicle — it is “degrades gracefully and recovers without a truck roll.” Dual SIM + modem watchdog + power failover is what makes teleoperation survivable across a whole shift of coverage changes, and it is why these features, not raw peak speed, are the ones to obsess over. |
One more practical note for teleop specifically: latency and jitter matter more than throughput. A 1080p stream at 6 Mbit/s is useless if it arrives in 2-second bursts; a 2 Mbit/s stream at steady 120 ms is usable. Size the carrier mix and the bonding against the interaction requirement (video + command + health), and validate it on the actual route, not in a coverage map.

06 · GNSS and RTK Delivery: Positioning the Fleet, Not the Pixel
Position is the one thing almost every vehicle deployment thinks it needs from the router and almost everyone over-states. The E-Lins in-vehicle platforms offer a GPS/GNSS option with a dedicated SMA antenna connector — the H820Q specifies an active GPS antenna on that port — and multi-constellation support (GPS, GNSS and BeiDou are documented on the M300 module). For an autonomous vehicle, that receiver does real, useful work: tagging telemetry with location, time-syncing nodes, and acting as the delivery channel for RTK correction data and map tiles from the fleet server.
| What the router’s GNSS is for: asset-grade positioning, time synchronization and delivering correction/map data to the vehicle’s own localization system. It is not the precise localization fusion the autonomy stack performs, and it should not be read as a positioning-accuracy claim. “GPS enabled” on a spec sheet tells you a receiver exists; it tells you nothing about the antenna, the cable loss, or the fix you will actually get on a roof under a gantry. |
The reason to be careful here is field experience, not pedantry. A “locked” GNSS fix can still be tens of metres off when the antenna is dash-mounted inside a metal cab, the cable run is too long with extra joints, or the connector expects an active antenna the installer paired with a passive one. For a fleet running between tall structures or under partial sky, multi-constellation reception genuinely helps; for open-route work it may not move the needle. Confirm the specific constellation support of your chosen SKU with E-Lins rather than assuming from a generic “GPS/GNSS” line — and mount the antenna with a clear sky view, short and direct.
| RTK the right way |
| When a deployment needs centimetre-grade positioning, the pattern is: a correction source (base station or NTRIP service) → the router’s cellular link → the vehicle’s localization unit. The router carries the correction stream; the autonomy stack consumes it. Keep those roles distinct, and the router’s GNSS becomes a useful secondary, not a single point of failure for position. |

07 · The Onboard Sensor Fabric: PoE, Serial and Digital I/O
An autonomous vehicle is a collection of subsystems that predate the autonomy — cameras, LiDAR edge nodes, legacy machine controllers, status contacts — and the router’s job is to give them one tidy, powered, addressable fabric instead of a tangle of dongles. Two capabilities do most of that work.
Power over Ethernet for onboard nodes
- The H900 and H900f support PoE PD or PSE up to 802.3bt (PoE++) — enough to power onboard cameras, a small edge-compute node or a wireless access point from the router’s own ports.
- That turns the router into the in-vehicle network spine: one device powers and connects the perception edge nodes, instead of a separate PoE switch and another power feed.
- Useful exactly where an autonomous vehicle already has a camera/LiDAR rack that needs both data and power run to it.
Serial and digital I/O for the legacy bus
- RS232 / RS485 with Modbus and DTU (serial-to-cellular) bridges older machine controllers and sensors that were never designed for IP.
- Digital I/O (DI/DO) on the H700/H750 range lets the router read status contacts and drive simple outputs — and on those platforms, I/O events can raise an SMS or Email alarm independent of the data tunnel.
- That last point matters: an alarm that leaves the vehicle over SMS when the data tunnel is down is the difference between “we found out at the next telemetry poll” and “we found out now.”
The throughline is consolidation. A vehicle that already carries a CAN/RS485 machine bus, a few IP cameras and a status panel does not need three vendors and four power bricks to get them online. One hardened router with PoE, serial and digital I/O is the fabric; the autonomy stack and the back-office each get the slice they need.

08 · Edge AI Offload: The Router Stages Data, It Does Not Pretend to Think
Autonomous vehicles generate enormous local data — camera frames, point clouds, vibration streams — and the temptation is to ship all of it. That is how a pilot turns into a connectivity bill that dwarfs the hardware. The discipline that prevents it is edge offload: do the heavy inference where the data is born, and let the router ship only the result.
| Honest boundary: E-Lins industrial routers are not GPU-class inference engines. They do not run the deep-learning models. What they do well is sit at the right point in the network to make edge offload work: reliable cellular, a VPN tunnel home, local staging, and the judgment to send only flagged events and summaries instead of raw streams. |
The architecture that actually holds up pairs a purpose-built edge compute device — running the real model — with the router handling everything connectivity-related. E-Lins platforms vary from a 580 MHz single-core / 512 MB unit (the H750) up to an 880 MHz dual-core / 4 GB unit (the H700), with the H900-series top variant configurable up to 8-core-class RAM/flash expansion. That is genuinely useful for local data staging and a lean on-router rules engine, but it is staging and triage, not training or inference at scale.
| Field pattern — offload, don’t stream |
| A vision-inspection pilot on autonomous yard vehicles started streaming raw frames to the cloud and watched the data cost blow past the connectivity budget in weeks. Re-architected: the companion compute runs the model locally, the router ships only anomaly events and a thumbnail over cellular. Same visibility, a fraction of the traffic, and the link became a reporting channel instead of a firehose. |
The takeaway for spec’ing: decide what leaves the vehicle before you size the link. If the answer is “raw sensor data,” the problem is architecture, not carrier choice — and the router is the device that enforces the boundary between local and remote.

09 · Fleet Security at Scale: You Cannot Harden What You Cannot See
Three hundred autonomous vehicles are three hundred internet-exposed routers the day they are commissioned — unless someone designs otherwise. At fleet scale, security stops being a per-device configuration exercise and becomes a visibility-and-process problem. The router features that matter are the ones that let you manage security across the fleet rather than hope for it.
| The lesson that sticks: in one fleet audit, 41 of 300 field routers still ran the factory default admin password — findable in the vendor’s own public manual. Centralized credential and firmware visibility turned a median patch time of 214 days into 12 days on the same hardware. The routers were never the weak point; the lack of fleet-wide visibility was. |
- Provisioning-time credential elimination. E-Lins’ Advanced Security Mode (local user management only) plus RADIUS/TACACS+ support lets a fleet centralize credential policy, and 802.1x adds LAN-side access control. Make rotation a mandatory, non-skippable provisioning step — before the unit ever reaches a vehicle.
- Certificate-based authentication. Removes the shared-secret problem entirely: no “default” to leave unchanged, and revoking one compromised device does not mean rotating a pre-shared key across the fleet.
- VPN breadth. IPsec, OpenVPN, WireGuard, DMVPN, ZeroTier, GRE/L2TP/PPTP available — match the protocol to each access path’s risk rather than defaulting. WireGuard and IPsec commonly sit under a zero-trust access layer rather than being replaced by it.
- Zone-based object firewall. Rules built around host, port and MAC objects support the granular access a zero-trust posture depends on, instead of flat allow/deny.
- OTA-managed firmware through NMS Cloud. A patching cadence is only real if you can execute it across the fleet without a truck roll to every site. Centralized push is what makes “patch this week” mean something.
For higher-risk paths — contractor access, internet-exposed management — layer a zero-trust access control on top of the VPN rather than assuming the tunnel alone is the perimeter. And above all: build the centralized inventory (firmware version, credential status, management exposure per unit) first. You cannot harden what you cannot see, and the autonomy on the vehicle does not change that.
10 · What to Spec: Matching a Platform to the Vehicle
There is no single “autonomous vehicle router.” The right platform follows the vehicle’s power, its onboard load, and how exposed the site is. The table below maps the E-Lins in-vehicle-relevant families to the role each fits; confirm the exact SKU and its constellation/POE options with E-Lins before procurement.
| Platform family | Vehicle fit | Standout for AV use | Watch |
|---|---|---|---|
| H820Q | In-vehicle, general | Dual SIM, ignition sensing, ISO 7637-2, active-GPS SMA option, 5–60V DC with dual power failover | IP30 metal case — install inside a vehicle cabinet or sealed enclosure |
| H820QO | Fixed outdoor / exposed site | IP68 outdoor enclosure, pole/wall mount, built-in high-gain antennas | H820QO for weather-exposed, cabinet-free mounts |
| H750 | Legacy-machine vehicles | Dual/tri power, RS232/RS485 + Modbus, DI/DO, ignition sensing, 5–60V | Entry compute (580 MHz / 512 MB) — staging, not inference |
| H700 | Heavier local staging | Dual-core, up to 4 GB RAM, DI/DO events via SMS/Email, serial | Use for on-router rules, not ML models |
| H900 | Camera/LiDAR-heavy rigs | PoE++ (802.3bt) to power onboard nodes, high compute headroom | IP30 metal case — enclosure still your job indoors |
| Selection rule of thumb: pick for power and environment first (ignition sensing, 5–60V, ISO 7637-2, IP rating), then for onboard load (PoE++ if you power cameras/nodes, serial/DI-DO if you bridge legacy kit), and only last for compute. Most autonomous-vehicle connectivity needs are met by the H820Q / H750 / H900 family. |
11 · Three Deployment Patterns
The principles above land differently by vehicle type. Three patterns cover most of what we are asked about.
| Pattern A · Autonomous mining truck |
| Large vehicle, harsh dust and vibration, frequently off-grid-ish power, and — critically — potentially a hazardous (gassy) zone at the face. Router in the non-hazardous cab/control bay, ignition-switched, dual SIM for pit coverage, serial/DI-DO bridging the existing machine bus, and an SMS alarm path independent of the data tunnel. The autonomy stack is local and safety-rated; the router is operations and supervision only. |
| Pattern B · Port or yard AGV / shuttle |
| Dense metal environment, constant coverage shadows from cranes and containers, and a strong teleop/supervision need. Dual SIM failover and the modem watchdog earn their keep here; PoE++ powers the onboard cameras from the router. RTK corrections delivered over the link keep positioning tight between the fixed references of a mapped yard. |
| Pattern C · Agricultural robot / autonomous machinery |
| Open-sky routes (good GNSS) but long distances from any backhaul, and seasonal, often unsheltered deployment. Ignition sensing preserves the tractor battery between shifts, multi-constellation GNSS helps at field edges, and edge offload keeps the link a reporting channel rather than a raw-stream firehose across expensive rural cellular. |

12 · FAQ
Q1:Can an E-Lins router drive the vehicle or run the autonomy software?
No, and this is the article’s central boundary. E-Lins industrial routers provide the connectivity layer — cellular backhaul, VPN, local bridging of onboard buses, positioning receiver and fleet management. They do not run perception, planning or control, and a design that routes safety-critical actuation through the router or the cellular link is mis-architected. The autonomy stack must remain local and independently safe.
Q2:What happens to the vehicle when the cellular link drops?
By design, nothing unsafe. The autonomy continues under its own local safety envelope — a controlled stop or safe handover — and the router simply stops being able to reach the outside world until the link recovers. That is why dual-SIM failover, a modem watchdog and power failover are specified: to make the drop rare, while the fail-safe behaviour makes it harmless when it happens.
Q3:Is the router explosion-proof for hazardous mining or petrochemical sites?
No. E-Lins industrial routers are not explosion-proof and carry no ATEX/IECEx certification. In a hazardous (classified) zone, the router must be installed in the non-hazardous compartment or a non-classified enclosure, with any in-zone sensing handled by certified explosion-proof equipment. We will say plainly when a site is outside what this hardware is rated for.
Q4:How accurate is the built-in GNSS, and can it be the vehicle’s localization?
The router’s GPS/GNSS option is for asset-grade positioning, time sync and delivering RTK corrections — not for the precise localization fusion the autonomy stack performs. Published accuracy figures describe open-sky reference conditions; real accuracy is set by antenna placement, cable loss and sky view. Confirm exact constellation support per SKU with E-Lins, and never read the router’s GNSS as a positioning-accuracy guarantee.
Q5:How do I keep a fleet of these secure without a truck roll to every unit?
Provision credentials during manufacturing (Advanced Security Mode + RADIUS/TACACS+), use certificate-based auth where possible, and manage firmware and visibility centrally through NMS Cloud with OTA push. The biggest lever is the centralized inventory: if you cannot say which units are on old firmware or default credentials right now, you do not yet have a hardening program.
| Conclusion: Build the Network Behind the Wheel, Not in Front of It |
| An autonomous vehicle is judged not by how clever its perception is on demo day, but by whether three hundred of them stay safe, updatable and accountable across a year of real routes. The connectivity layer is the quiet part of that story — and, done right, it is the part that makes the rest manageable. |
| Three things to settle before you specify connectivity for an autonomous or unmanned vehicle: —Keep the link out of the safety path. Decide, with the vehicle’s safety authority, that the autonomy stack is local and independently safe, and that the router is operations-and-supervision only. This is the one boundary on this list that is a safety issue rather than a cost issue, and it is where the whole design has to start. —Specify for the vehicle’s electrical reality, not the bench. Ignition sensing that does not drain a parked battery, 5–60V input with dual/tri power failover, and ISO 7637-2 transient protection are what let a router actually live in a vehicle instead of failing in month two. The hazardous-area question — where this hardware is, and is not, rated to go — gets answered with the safety authority, not after a failure. —Treat the link, the alert path and the fleet security as one managed system. Dual-SIM failover and a modem watchdog keep teleoperation alive; an SMS alarm that leaves the vehicle when the tunnel is down keeps you informed; and centralized credential and firmware visibility through NMS is what turns “we think the fleet is patched” into “we know it is.” The router makes all three possible; it does not do them for you. |
Building Connectivity for Autonomous or Unmanned Vehicles?
Tell E-Lins your vehicle types, their power and onboard-load environment, and where any classified hazardous area sits. We’ll confirm the ignition sensing, the 5–60V and ISO 7637-2 protection, the GNSS and PoE options, and the right platform — H820Q / H820QO, H750, H700, H900 / H900f — for the network you actually operate. For hazardous areas, we will say plainly that the equipment is not ours to supply in-zone.






