about
Moravec combines industrial operations, remote control and machine learning to make existing heavy machinery more productive. We start by delivering work at lower cost, then use that experience to build autonomous capability.
mission
Our goal is heavy machinery that does useful work on its own, with people supervising and handling exceptions. Remote operation gives customers value now and supplies the experience our models need to learn.
available now
in pilot
in pilot
in development
the founders
Co-founder & CEO
Twenty years building automation, computer vision and managing some of the world's largest automated industrial aquaculture operations across Australia, the US and South East Asia.
Leads Moravec’s commercial strategy and deployment into working sites.
Co-founder & CTO
Machine learning and robotics engineer who has shipped production AI at scaling startups, at Microsoft and across large government data programmes.
Leads the technology that connects remote operation, learning and autonomy.
why the name
In 1988 the roboticist Hans Moravec named the strangest fact in artificial intelligence: reasoning, the ability we prize, is cheap for computers, and the physical skill we take for granted is the hardest thing a machine can learn. Chess fell decades ago; loading a truck still has not.
His explanation still stands. Evolution spent half a billion years refining perception and movement and only a moment on arithmetic, so the work that feels effortless from a cab is the deepest computation on earth.
Language models learned from an internet the world spent thirty years writing. No such record exists for skilled machine operation, so we are recording one: every command, every camera, every outcome, on working machines.
The recordings that matter most are the edge cases: the tangled grab, the bank that will not release the bucket, the load that shifts mid-swing. A simulator does not write those, and a lab cannot stage them.
An ordinary week of production writes them by the dozen, each with a skilled operator's answer attached. That is the data that teaches a machine, and the name of this company is a promise to collect it.
Hans Moravec with the Stanford Cart. In 1979 it crossed a chair-filled room on its own. It took five hours.
how we work
01
Begin with an agreed workload and a clear commercial case.
02
Compare performance with the site’s baseline.
03
Assess and commission each deployment.
04
Validate autonomous capability before expanding its use.
the direction of travel
More machine hours per operator, not fewer operators per machine.
where we are