Engineering directions in design or early build. Published because partners who care about these problems should help shape them — not because you can buy them today. Nothing on this page is a product capability.
Mesh-capable radios are expensive, and multi-platform meshes often mean multiple radios per aircraft. The direction: the overlay computes a GPS-synchronised transmit schedule so one half-duplex commodity radio per node serves the whole neighbourhood — moving the intelligence from the radio into software, and eventually supporting alternative time sources for GNSS-denied operation. Design phase, published on the engineering site.
The elevation-model line-of-sight and Fresnel-zone engine is built, unit-tested, and already drives the mission simulator — every link colored by real terrain geometry. The remaining work is wiring its risk output into live scheduling and multi-hop routing, alongside range and signal risk, on one shared cost model. See it in Atlas Sim
Terrain awareness teaches the network where geography breaks a link; the same mechanism can carry the threat picture. Zones declared contested — an electronic-warfare overlay from mission planning, a reported jamming area, an emission-control boundary — load like map tiles, each with the frequency bands it affects and the hours it applies. A platform heading into a declared zone sees the risk on its affected links rise before the first packet is lost there, redundancy escalates in advance, and links on other bands carry on unpenalised. Farther out, the fleet’s own shared telemetry can corroborate where the spectrum has actually been hostile. Advisory by the same principle as every prediction source: zones raise costs; live measurement decides the truth.
With positions and velocities in the routing flood, the network can compute where a mobile relay should be: “the ground station and the forward pair lose line-of-sight behind the ridge in about 40 seconds — a relay at X prevents it.” Advisory output to the C2 system; never an automatic flight command.
Adapters already carry radio telemetry upward: signal strength, noise floor, buffer depth. The same boundary can carry intent downward. Atlas holds the only cross-vendor view of the whole topology and the mission’s objectives — the traffic classes each segment is carrying right now — so it can tell a segment what “good” currently means: optimise for latency while the command corridor runs through you; optimise for throughput while you carry the sensor backhaul; optimise for energy when only patient traffic remains. With position and velocity in the routing flood, intent can even arrive before the need: “the forward pair enters the ridge shadow in 40 seconds — prefer low-latency forwarding.”
Objectives, never knobs. Atlas would never set channels or data rates — the radio always owns its own physics. Intent is advisory, delivered slowly with hysteresis through vendor-supported interfaces, authenticated and audited — the same design principle the shipped prediction engine follows: advise, never override. For radio vendors this is the partnership surface: the radio stays the expert; Atlas makes it mission-aware.
Candidate strategies under evaluation: cost-aware placement for metered links, minimum-jitter selection for industrial control, loss-weighted distribution, sub-packet spraying, and explicit active/standby with recovery hold-down.
Transparent compression between fragmentation and encryption — high value for plaintext-heavy traffic such as SCADA and legacy radio protocols, with a conditional send-uncompressed path so incompressible data pays nothing.
Research sequencing is negotiable — the directions below are prioritised by real operational demand, and a briefing is where that demand gets registered.
