01
Runs with the network off
Detection, tracking and fusion run on the node. No data centre, no cloud, no link required.
Software
EdgeOS
The operating layer for systems that must decide without the network.
EdgeOS runs on the node, turns raw sensor streams into tracks and events, and keeps working when the link is jammed or gone. It sends only what matters, when it is allowed to.
Sensors multiply every year, and the link back is the first thing to fail. Streaming raw video to a data centre assumes a network the field rarely has. EdgeOS does the work where the data is, and sends only what matters.
How it works
From feed to decision,
in four steps.
- 01ConnectLink the node to the sensors already in place: cameras of every kind, drone video with its position data, radar and lidar, and more through adapters. No new sensors needed.
- 02DetectAI models run on the node itself and mark what they find in every feed, each find with a place on the map.
- 03CombineBuilt to recognise one person or vehicle, seen by several sensors, as one track.
- 04DecideThe right people see what needs a look, on site or in the systems they already use, and decide what happens next.
What the models find
Many sensors.
One picture.
Four angles on one place: fixed cameras and the site's own drones. EdgeOS is built to know it is the same person in every view, keep following them, and tell the people on site.
P2P3P1VH1
P2P3P1VH1
P2P3P1VH1D1 Own droneD2 Own drone
P2P3P1VH1
P4P5VH2VS1VS2VS3
P4P5VH2VS1VS2VS3
P4P5VH2VS1VS2VS3D3 Own droneD4 Own drone
P4P5VH2VS1VS2VS3
P6P7P8VH3
P6P7P8VH3D5 Own droneD6 Own drone
P8P7P6VH3
P6P7P8VH3Illustration only, not the EdgeOS interface. Generated images: no real sites, people or decisions. Matching one object across sensors and flagging it are what EdgeOS is built to do.
Capabilities
What makes it different.
Any sensor
Anything that reports a measurement can be fused, one adapter per sensor type. Cameras of every kind, drone video, radar and lidar today.
Meaning, not video
EdgeOS sends tracks and events, not streams: more than 99.9% less radio traffic than four HD camera feeds (modelled).
Nothing lost
If the link drops, results wait on the node and go when it returns. Zero events lost across 77 days of field data.
Quiet by policy
Emission control in the software: the unit decides what may leave the node, when, and over which link.
Into the systems you use
Tracks and events go into Palantir Foundry today. Built to feed other C2 and mission systems through adapters.
Where EdgeOS goes
Built on what
runs today.
- 01On the nodeTodayDetection, tracking and fusion on the node, quiet by policy, nothing lost when the link drops.
- 02Our own foundation modelBuildingOne model across cameras, drone video and other sensors that keeps one identity per object, judges what is worth sending and knows when it is unsure. Developed and trained in-house.
- 03Agents on the nodeNextAgents that reason over the fused picture and work through a fixed set of tools: task a sensor, cue another node, draft a report for the operator.
- 04Assured autonomyDirectionThe commander sets how far a system may act. Limits live in code outside the model, and every decision is recorded. Today, EdgeOS advises and people decide.
In numbers
Measured, not promised.
79ms
Detection on the node to Palantir Foundry
Measured232ms
Camera glass to the operator's screen
Measured0
Events lost across 77 days of field data
Verified> 99.9%
Less radio traffic than four HD camera streams
ModelledSpecifications
01 / 05
Runs on
EdgeOS is hardware-agnostic: it runs on our own nodes, HP-FN1 and HP-FN1 Micro, or on the edge compute a platform already carries.
- Nodes
- HP-FN1, HP-FN1 Micro Product line
- Compute
- Any capable edge compute; our nodes up to NVIDIA Jetson Thor (2070 TFLOPS) Today
02 / 05
Inputs
Links to the sensors already in place.
- Inputs
- Cameras (IP, GMSL, infrared and thermal), drone video with MISB ST 0601 KLV, radar, lidar, GNSS and IMU Today
- New sensors
- One adapter per sensor type Design
03 / 05
On the node
AI models run on the node itself, on every connected feed.
- Models
- Detection and tracking on video; detections fused into tracks and events Today
- Camera to picture
- 232 ms Measured
04 / 05
The picture
One picture from every sensor, and what is happening in it.
- Tracks
- One track per object, on the map, however many sensors see it Today
- Events and incidents
- Rules on the node: what, where, for how long, how serious Today
- Meaning
- What is happening, not only what is there. Today in Palantir Foundry; moving onto the node. Building
- C2 and mission systems
- Into the C2 and mission systems the unit already runs, one adapter each. Palantir Foundry live today. Today
05 / 05
On site
Keeps running with no outside link, and decides what leaves the node.
- Outside link
- Not required; store-and-forward when there is one Verified
- Emission control
- Set per link by the unit Today
01Does EdgeOS act on its own?
Not today. EdgeOS shows people what it finds, and people decide what happens next. Where autonomy goes further, the commander sets its limits, mission by mission.
02Does the node come with sensors?
No. It connects to the cameras and sensors already in place.
03Does it need a network?
No. Everything runs on the node. When there is a link, EdgeOS uses it within the limits the unit sets. When the link drops, results wait on the node and go when it returns.
04Where does the data go?
Only where the unit allows. The unit sets what may leave the node, when and over which link. Tracks and events are what travel.
05Which sensors does it work with?
Cameras of every kind: IP, GMSL, infrared and thermal. Drone video with KLV, radar and lidar, with GNSS and IMU for position. Ask us about yours.
06Can several nodes work together?
Yes. Every node publishes into the same picture in Palantir Foundry today. Nodes sharing one picture directly, with no central system, is what we are building now.
