Skip to content

Edge AI · Physical AI · Embedded systems

Decisions made at the Edge.

We build physical AI, edge AI and embedded systems for solving problems that today’s infrastructure creates. We carry deep research into technology that produces inference which runs where the data is made, on the hardware that has to survive the room it sits in.

  • THE LAB. We take the problem nobody has solved. Applied research, ending in hardware you can buy
  • DRONADEEP. AI payloads for drones. Edge compute and sensing, carried by the aircraft
  • IN FIELD TRIALS. Navigation when the signal is gone. Dronadeep’s GNSS-denied navigation box, in trials
  • EDGEBOXX. Research, packaged as product. Applied R&D, built into deployable boxes
  • EDGENTRIC LABS VISION. Embedding Edge Intelligence. Edge AI and embedded systems for the physical world.

What we make

Edge AI products and Solutions

We pick client problems in our applied research track. A client looking to solve a specific practical problem, who has not found an answer anywhere. Our research team finds the technology that solves it. And in due course, many of these turn out to solve repeat problems across various industries. When that happens, these solutions become applications you can buy pre-configured - Dronadeep, our UAV companion box, is the first such product.

  • EdgeBoxx

    The Multi-faceted Box

    An edge compute box that arrives configured for a task rather than as a general-purpose computer you have to make useful. You tell us the problem, we ship the box with the connectivity that suits the site and models tuned to your process. We choose the configuration of the box that fits your size, bill and use-case.

  • Dronadeep

    The airborne variant

    Built on the same platform, in a form an airframe can carry, for drone manufacturers who want the aircraft to be intelligent at the point of manufacture rather than after it. GNSS denied navigation, auto-pilot, drone health, target tracking and indoor navigation are the use cases it targets.

  • Applied R&D

    Our Research work

    A dedicated team that takes on problems nobody has a product for yet. Scope, build, trial, deploy, and stay reachable afterwards. Some of that work stays bespoke for the client. Some of it becomes the next thing on our list of products.

Work

Research and problems we work on

We get problems from all sides that require solutions by embedding intelligence in existing machines. A panel that has no data port but must read a 7-segment display and send to the server. An imported angular sensor that needs a nine-month lead time to build. A measurement somebody still writes on a clipboard instead of capturing automatically. Three problems which have nothing to do with each other but have one thing in common - they all need a solution without replacing the machines themselves. Different industries, different equipment, different people asking for a solution without needing an overhaul. And that is the point of what we do at Edgentric Labs. What carries across is not the problem we initially see, it is the compute and the sensing underneath it that plugs into an existing system. We don't build Drones, but we make existing ones smarter.

  • Bringing an aircraft home when GNSS is gone

    Defence, survey and commercial UAS

    Navigation in a GNSS-denied environment is a problem traditional IMUs suffer with, and most return-to-home behaviour treats the loss as a single event with a single response.

    We are building a retrofit module that treats it as a ladder of degraded states instead. It fuses a quantum MEMS magnetometer and IMU, a fibre-optic gyroscope, mmWave radar and a world magnetic map, records the outbound path including the highest point reached, and sits at the junction between the flight controller and the ESCs so that it can fly the aircraft home when the link is gone rather than only advising. Every component in the set is non-ITAR, and sourcing is constrained by what these programmes will accept, which for these customers decides whether the module can be fitted at all.

    Written up in full as a research note.

  • Running a battery swapping station from one Android panel

    EV energy infrastructure

    A swapping station has to know who is standing in front of it before it opens anything. Which driver, which cabinet, which battery, and whether that battery is in a state worth handing over.

    We built the panel that answers all of it. A 7 inch Android HMI on a Qualcomm system-on-module, with an NFC reader for the driver's card, driver identity and access management, and a live link down to the station controller for reading the state of every battery slot and for opening the one the driver is owed.

    And the panel is not only a screen. The module carries enough compute to run models on the glass itself - face detection at the station is the first of them, which means a driver can be recognised without a round trip to anybody's server.

  • Reading a panel that has no data port

    Battery systems, energy

    A battery-health system needed readings from equipment that had no data port to give them. The values existed only as digits on a 7-segment panel, read by eye by somebody walking over to look at it.

    We put a box in front of the display. It recognises the digits on-board and puts them onto the CAN bus as structured values the health system already understands. The equipment is untouched, the monitoring system did not have to change, and numbers that used to be collected on foot are now continuous machine data.

The full body of work →

What carries across

Embedded Intelligence on a Machine

A standard battery panel. A regular Drone. A shaft angle on naval equipment. A ship's bridge. For every problem we see, the same three constraints turned up every time. Automate, embed intelligence and ship to retrofit an existing setup.

  • Retrofit before replacement

    Do you have a fleet already flying? Instruments already fitted? A line already running? We see this all the time. Almost every one of these problems arrived attached to equipment which nobody was going to replace. And replacing any of it is a capital decision that takes years, not a purchase somebody signs off in a quarter. So the only useful answer is not a new machine, it is capability you can add to what is already there. And the trade-off? It is a harder engineering constraint for us - the capability has to work on equipment we did not design. And a far easier proposition for you to deploy.

  • A companion box, fluent both ways

    The shape that kept appearing is a companion box - it sits between two systems and is fluent in both. On one side it reads whatever the equipment actually produces. Digits on a panel. A shaft angle. A sensor sentence. A flight controller's heartbeat. On the other side it puts that output back in the format the incumbent system already understands. CAN for the battery monitor. Modbus for the motor controller. MAVLink for the aircraft. IEC 61162 for the bridge. So nothing downstream has to change. And that is usually the difference between a project which gets approved and one which does not.

  • It has to work when the network does not

    None of this runs on a network you can rely on. It is unreliable, it is contested, or it is not there at all. That is why the decision happens on the device and not after a round trip. Why the mesh work has no node whose loss takes the network with it. Why the aircraft can fly itself home. And why anything we build can be updated without somebody physically reaching it. That is EdgeBoxx. Not a strategy we wrote down and then applied to things, but the shape the work kept taking.

Next step

Put it on your hardware.

We run scoped technical evaluations with integrators, OEMs and operators. Tell us the platform and the constraint you are up against, and we will tell you whether we can help.