Book a call

Melbourne, Australia

Trusted compute for physical AI

Enoxa designs the safety-critical, real-time control layer that sits beside the AI brain of any robot or autonomous machine. We start with robots and a certified module, and build toward our own RISC-V chips, designed in Melbourne.

Trust engine · audit logIllustration

14:02:07.118 teleop.move v=2.4 m/s  signed ✓ policy ✓

             → clamped to 1.2 m/s · person at 3.1 m

14:02:09.540 fw.update v4.2  unsigned · dropped

14:02:11.003 video.upload site-cam-2  egress denied

1kHzControl loop, under 10 µs jitter
SIL 2IEC 61508 functional-safety target
<5WE1 power, for battery and field machines
22nmAt an allied, non-Chinese foundry

E1 design targets, not measured results.

The problem

Robots have a brain. Their cerebellum is a patchwork.

Every capable robot runs two kinds of compute. The brain runs large AI models for vision and planning. The cerebellum runs motion control, sensor fusion and safety stops, where milliseconds matter and every part must be certified.

Today that second job is handled by general-purpose controllers that were never designed for whole robots. That is the layer Enoxa builds.

AI brain: one 30 Hz camera frame ≈ 33 ms Cerebellum: 33 control cycles in the same time, 1 ms each Enoxa design target: 1 kHz loop
BrainCerebellum (Enoxa)
JobSee, understand, planMove, sense, stop safely
TimingTens of milliseconds is fineEvery millisecond, on a fixed schedule
SafetyNot safety-ratedDesigned for IEC 61508 SIL 2
ChangesWhenever AI models improveDesigned in once, certified, kept for years
TodayNVIDIA Jetson Thor and similarA patchwork of general-purpose controllers
How Enoxa works

The brain decides what the machine should do. Enoxa decides whether it is allowed and safe.

Three gaps we close

Trust

Robots and autonomous machines in mining, defence and critical infrastructure need control hardware their operators can audit and trust.

Fit

Robots with 30 or more joints and learned AI control need more than a controller built for a single machine.

Sovereignty

Australia designs world-class chips. Its robots and autonomous machines should run on compute designed here.

One path from module to chip

Each step funds the next, and a robot designed around M1 takes E1 later without a redesign.

M1Coming 2028

Control module

Our safety software, robot interfaces and a link to the robot's brain on a pre-certified safety chip.

E1In development

Cerebellum chip

Our own RISC-V chip, with real-time cores, a robot AI engine and a dedicated safety island.

E2On our roadmap

Brain and cerebellum

A full chip with HBM memory for allied makers of robots and autonomous machines.

See the technology

Where it runs

Built for machines that can't fail

SegmentWhat Enoxa does in the machineProduct
Mining autonomySpeed and separation near people, signed remote commands, machine-health models for haul trucks and drill rigsM1 → E1
DefenceMission-command authorisation, a tamper-evident audit log, air-gapped fleet toolsE1
Agriculture and fieldPerson detection near implements, rugged low-power control far from the gridE1
Robot and vehicle makersA certifiable, AI-ready controller beside the robot's brainE1 → E2

Built with the people who run robots

We develop every product inside real robot fleets, starting with ARIBOT. If you build or run robots or autonomous machines, help shape what we build.

Join the design-partner program