Skip to content
Sentia
HOW IT WORKS

A loop built to out-pace the emergency.

SentiaCore's reasoning runs as one continuous loop — sense, recreate, reason, rehearse, respond, and learn from it — fast enough that the plan a robot is executing was built on a picture of the scene that's still true.

THE LOOP, END TO END

Six moves, repeating.

Every mission, from a single sensor alert to a multi-robot response, moves through the same six stages. Nothing skips the loop — not a doctrine rule, not a language instruction, not an edge case nobody anticipated.

  1. 1

    Perception

    Every robot's cameras and thermal sensors feed a shared read of the scene — what is actually visible right now, nothing assumed.

  2. 2

    Recreate

    A living twin of the scene, built only from what the team has sensed — where fire and hazard are confirmed, where they are forecast, and where the picture is still uncertain.

  3. 3

    Reason

    The team weighs its options as one mind. Every robot bids on the work in front of it; the strongest candidate plans rise to a human commander for the call.

  4. 4

    Rehearse

    Before a single robot moves, a plan is run forward against the twin — played out next to doing nothing, next to the alternatives — and scored on what it actually changes.

  5. 5

    Respond

    The winning plan moves into the world through whichever bodies are on site — ground robot, legged robot, drone — each holding its own reflexes and a safety shield that a human can override at any instant.

  6. 6

    Self-improve

    Every mission becomes material: replayed, searched for the edge case that nearly broke the plan, and turned into a new scenario the team rehearses against next time.

One continuous loop. Not a pipeline that ends — a habit the team repeats faster than the emergency changes.

REHEARSAL, NOT A GUESS

The plan is tested before it's trusted.

Before SentiaCore lets a plan move a robot, it forks the living twin and runs the plan forward — then runs the alternative (often: doing nothing) forward on the same starting conditions. The two outcomes are compared on what actually changes: how much more area stays unburned, how many more people are reached, how much risk is avoided. A plan has to earn the win, not just sound reasonable.

  • The comparison always includes a baseline — a plan is judged against an alternative, never in isolation.
  • Rehearsal runs on the same physics the team is actually facing, not a simplified stand-in.
  • A plan that only looks good in theory doesn't survive the rehearsal.
  • In one rehearsed scenario, holding robots back to search for people before attacking the fire cost more than twice the burned area within three minutes of the decision — and past four times the area within five, a gap that kept growing rather than leveling off.
A BELIEF THAT KNOWS ITS OWN LIMITS

The twin tracks what's confirmed, forecast, and unknown — separately.

SentiaCore's picture of a scene is not one layer of truth. A hazard that's been directly sensed, a hazard the physics model forecasts will spread there next, and a space nobody has looked at yet are three different kinds of knowledge — and the system never lets the second or third stand in for the first when a robot's safety depends on it.

  • A life sign confirmed once stays confirmed — it is never silently retracted for lack of a recheck.
  • A space nobody has sensed is marked unknown, not assumed clear.
  • Confirmation ages: a reading from a while ago carries less weight than one from just now.
THE SAFETY GATE

Every action clears the same fixed set of rules.

Doctrine, the planner, and any language-driven proposal all answer to one deterministic gate before a robot acts on them. The gate holds no model and no randomness — only rules, each one added because a real situation demanded it.

Rules are deterministic.
The gate's decision doesn't depend on a model's mood or a random seed — the same situation gets the same answer.
The reason is always legible.
When the gate blocks a proposal, it says why, in terms a commander can act on.
The override sits outside reasoning.
The emergency stop is wired beneath this entire loop — it works even in a failure mode the loop itself never saw coming.
Measured in rehearsal, not just claimed.
In one rehearsed run, every robot in the fleet came to a stop within a single control tick of the command — a tenth of a second.

See the loop run on a real scenario.

Operation Alpha walks through all six stages on one incident, start to finish.

Request a demo