A Volantyx Aerospace company
volantyx.com →
A Volantyx Aerospace company
volantyx.com →
Neurex is the autonomy runtime that sits between your AI and your actuators. It verifies every decision before it becomes physical action — so your robots can operate safely in unpredictable environments without constant human oversight.
Model-agnostic. Hardware-agnostic. Works with ROS 2, NVIDIA Isaac, and OEM stacks.
“Your robot works in the lab. It fails in the field.”
Novel situations, sensor degradation, unexpected humans, edge cases your training never covered. Capability doesn't generalize — and one surprise is all it takes.
“Regulators want evidence, not promises.”
The EU AI Act, ISO 10218, and ISO 3691-4 ask for traceable decision records and demonstrable safety architecture. Testing alone doesn't produce that evidence — architecture does.
“One incident kills your deployment.”
A single autonomous failure can shut down an entire fleet. Insurance, liability, customer trust — all of it rides on your worst decision, not your average one.
The bottleneck isn't capability. It's trust.
Neurex doesn't replace your AI. It governs what reaches the physical world.
Every robot decision passes through deterministic admission checks before reaching actuators — grammar, bounds, authority, resource feasibility, proximity, contracts — in real time.
A runtime assurance layer monitors execution continuously and intervenes on any constraint violation — even if the primary AI stack fails completely. Assurance is independent from intelligence.
Every consequential action generates a tamper-evident record: what the robot perceived, believed, decided, verified, executed — and what happened next.
Your foundation model — NVIDIA, OpenAI, proprietary, classical planner, or RL policy — perceives, reasons, and plans. Neurex governs the transition from inference to action: behaviors are declared with explicit preconditions, constraints, fallbacks, and authority levels, then verified before execution.
WORKS WITH ROS 2 · NVIDIA Isaac · MoveIt · Nav2 · custom stacks
HARDWARE-AGNOSTIC · MODEL-AGNOSTIC
Neurex implements the eight layers of the DPI reference architecture. Intelligence is separated from authority: models recommend, the architecture decides what may act.
Retrieval before synthesis. Verification before action. Assurance independent from intelligence.
Every deployment adds verified, evidence-backed behaviors to a growing library. Each behavior carries its operational history — what worked, what failed, under what conditions, with what confidence. The library compounds into institutional memory for physical AI.
EVERY BEHAVIOR CARRIES specification · constraints · assumptions · verification artifacts · hardware requirements · operational evidence · version history · failure modes
Fleets improve together — without sharing raw operational data.
Deploy the Neurex audit layer on any robot — even without the full verification stack. Get complete decision records for every autonomous action, and satisfy regulators, insurers, and customers with evidence instead of promises.
Start here. Add verification and assurance when you're ready. It's the fastest path to value.
Foundation models are now capable enough to plan physical actions — but ship with no safety governance between inference and actuation.
The EU AI Act (Articles 12, 14, 15), ISO 10218, ISO 3691-4, and FDA AI/ML guidance increasingly require architectural evidence — not just testing.
Warehouse AMRs, security robots, autonomous vehicles, humanoids — the fleets are coming, and trust is the deployment bottleneck.
Two published papers define the stack: the DPI reference architecture — the vendor-neutral framework Neurex implements — and the Verified Behavior Library, the evidence-centered assurance architecture behind the library, from behavior objects and the verification gate to evidence maturity and the assurance graph.
We're working with select partners to validate Neurex across security robots, warehouse AMRs, and industrial mobile platforms. Limited spots available.