A reasoning engine for the physical world

• Governed Knowledge: What the engine knows decides how it can act
• Perception: Observations with stated confidence
• World view and spatial intelligence: An operational understanding of the world
• Memory and learning: Memory gives every decision context
• Reasoning and decisions: The same knowledge and the same inputs give the same decision
• Coordination and execution: Independent platforms, one shared mission
• Ownship: the self-aware platform: Knowing its own capabilities and limits

Governed Knowledge: in more detail

Neural perception turns sensor data into observations with stated confidence. A deterministic symbolic core reasons over them with Governed Knowledge: domain ontologies, policies and operational experience that experts have approved. Policies are part of that knowledge, so they shape every decision from the inside.

Domain ontologies

The vocabulary of the operating world: which things exist, what properties they have and how they relate. In the recording: surfaces, the drone’s pose and five maneuvers.

Policies

Rules about what may be done, written in the same language as the rest of the knowledge. In the recording: keep 0.95 m beyond a surface’s radius and uncertainty, and never treat unseen space as free. A path that breaks a policy is not chosen, because the reasoning itself excludes it.

Experience

Models calibrated from earlier runs: command delay, response lag and how fast the uncertainty of an unseen surface grows.

Versions

Experts review and approve a version before it is used. Each decision records the version it used, so it can be replayed exactly. Live observations update memory, not the approved knowledge.

Perception: in more detail

Perception supplies observations; the deterministic core reasons over them. Observations are always read within the sensors’ coverage and limits, and uncertainty stays visible instead of being guessed away.

In the recording, a drone in a flight simulation watches the scene ahead. A neuron model that responds to approaching surfaces shows where something is growing in the image, sector by sector.

World view and spatial intelligence: in more detail

Spatial intelligence works on that picture: locations, relationships, visibility and physical constraints, and possible future states within a defined time horizon.

Digital twin

The commissioned reference model of the operating environment: its spaces, objects, infrastructure and relationships. It gives observations a known structure and location.

In the recording, surfaces seen by one camera are placed on a top view from parallax, and candidate paths are checked against them.

Memory and learning: in more detail

Short-term memory

Recent observations, active objectives, current plans and immediate context. It updates as new evidence arrives.

Long-term memory

Governed Knowledge, models of the platform and its world, and relevant history.

Fast and slow thinking

Fast thinking recognizes familiar situations and applies established procedures. Slow thinking examines alternatives, projects consequences and plans. Both draw on the same memory.

Learning is deterministic and governed: induction proposes rules from examples, abduction proposes explanations and knowledge repairs, and experts review each proposal. Only an approved, signed version becomes part of Governed Knowledge.

Reasoning and decisions: in more detail

01 Snapshot

Bind observations, objectives and relevant memory into one time-bounded picture.

02 Spatial intelligence

Locations, relationships, visibility and physical constraints.

03 Prediction

Possible future states within supported models and a defined horizon.

04 Candidate actions

Only actions that the platform’s capabilities and the policies in Governed Knowledge allow are derived.

05 Evaluation

Candidates are compared against objectives, predicted consequences and operating constraints.

06 Action

The selected action is issued with its explanation and decision record. Feedback informs the next cycle.

Policies are not a checkpoint after the decision. Where a policy requires a person to confirm a class of actions, the engine asks for that confirmation as part of the same reasoning. The engine never holds the controls: the platform’s own controllers and safety systems execute the decided intent, and every cycle can be replayed bit for bit.

Coordination and execution: in more detail

Shared mission

Each platform connects its observations to a shared understanding of the mission and its priorities.

Complementary roles

Work is divided by capability, position, resources and limits, weighed against completion time and energy use.

Coordinated execution

Routes, timing and dependencies are coordinated; progress and changes are shared so the team can redistribute work.

Execution stays with each platform’s own systems: flight stabilization and control are left to the aircraft. In the recording, a fast visual reflex turns the approach signal into a turn and a brake, and the simulated drone passes beside a building.

Ownship: the self-aware platform: in more detail

State

What is true right now, as one time-bounded picture.

Rules

What is permitted, required or forbidden, read from the policies in Governed Knowledge.

Capabilities

What the platform can physically do, bounded by its kinematics and limits.

Objectives

What the operation is trying to achieve and how competing aims are prioritized.