Karta InnovationsKarta Innovations
The autonomy stack

Three layers, one closed loop.

Intent and doctrine flow down through three layers; situation and audit flow back up. Each layer ships as a real product.

The architecture

Stack architecture

Intent and doctrine flow down. Situation and audit flow back up. Three layers, one closed loop.

Every action becomes the next situation

01 / 03On-vehicle · sub-50ms

L1 · Edge

SkyAnchor

Position the vehicle can trust without GPS, with a confidence interval on every fix.

The detail

The problem

GNSS is the single most fragile dependency in the autonomy stack. Jamming, urban canyons, and high-latitude operation all silence it. Without a substitute, the vehicle stops being useful.

The approach

SkyAnchor runs as a single deterministic process on-vehicle. There is no cloud, no model download, no online learning. The fusion is exposed as a service with explicit confidence intervals, so downstream consumers can reason about pose uncertainty instead of pretending it doesn't exist.

Capabilities

  • Visual-inertial odometry with multi-camera support
  • Terrain-relative navigation against on-board map tiles
  • Bounded drift on terrain-relative fix
  • Bounded behaviour as confidence degrades
  • Deterministic - no online learning in the flight loop

Integrations

  • Kari, Cerana, Dorsata
  • Third-party PX4-class autopilots
  • TENERE pose consumer
The SkyAnchor edge positioning module, a sealed sensor unit.L1 · Edge
02 / 03Operator station · ground

L2 · Command

TENERE

Bounded operator intent becomes a sortie graph, deconflicted across the fleet.

≤12

Per operator

Frame-level

Audit

200 ms

Latency

The detail

The problem

Operator cognitive bandwidth is finite. Naively, a fleet just multiplies the operator's workload. The solution is not to remove the operator. It is to compose mission intent at a higher level of abstraction and surface only the decisions that need a human.

The approach

TENERE is a constraint solver wrapped in an editor. Operators express what they want as bounded intent; TENERE deconflicts, plans, and runs the sortie graph. Every action is reversible. Every decision the operator did not see was provably out-of-scope of what they asked.

Capabilities

  • Mission graph composition with explicit bounds
  • Multi-vehicle deconfliction at composition time
  • After-action telemetry and decision audit at frame rate
  • Doctrine-aware planning surface operators can read and edit
  • One operator scales to a squadron of vehicles

Integrations

  • Karta Apis family
  • Kiksuya doctrine runtime
  • Standard military comms stacks
The TENERE mission composer over a live airfield.L2 · Command
03 / 03Squadron scale · regional

L3 · Theatre

Kiksuya

Doctrine, authority, and audit, remembered and provable across the theatre.

Squadron

Scale

Hierarchical

Policy

Federated

Memory

The detail

The problem

Doctrine in most autonomy stacks is a slide deck. It is not enforced by the code, it is not versioned, and it cannot be audited after the fact. That gap is where bad outcomes happen. The same goes for institutional memory across sorties - most stacks forget.

The approach

Kiksuya is a runtime that takes operator intent and a published doctrine, and produces an executable plan annotated with which doctrine clauses authorise which step. The annotation is logged at telemetry frame rate, federated across echelons, and survives the sortie. Doctrine becomes a property of the running plan, not a checkbox on a slide.

Capabilities

  • Doctrine encoded as versioned, signed policy
  • Per-action authority audit trail
  • Plan-level dry-run with explicit clause attribution
  • Federated memory across echelons and bases
  • Procurement-grade traceability from contract to flight

Integrations

  • TENERE planner
  • Karta Apis family
  • Customer-published doctrine packs
The Kiksuya theatre dashboard with doctrine zones and an audit log.L3 · Theatre

Integrate the autonomy layer into an airframe, platform, or programme.

Open an engagement