autonomy
Autonomy, trained
on real work.
Moravec is building autonomy for heavy machinery through productive remote operation.
Skilled operators complete the work today. Their actions and outcomes supply the training data for AI to perform more of it.
the learning loop
The work teaches
the machine.
We record how skilled operators respond to real tasks and changing conditions. We train models on that data and test them before any autonomous function enters an approved workflow.
Teleoperation creates the data. Physical AI turns it into autonomy.
We study each human intervention to find where the models need more training.
perform → record → train → validate → deploy
one progression
Four stages.
One gate each.
Customers can buy remote operation today. We develop and validate assistance and autonomy for specific tasks, machines and operating conditions.
-
01
Remote operation
available now
A skilled operator controls the machine and covers the agreed workflow.
the gateThe safe stop is commissioned and witnessed on site, and the operator drives as well from the station as from the cab.
-
02
Operator assistance
in pilot
Validated functions support selected movements while the operator directs the work.
the gateEach function beats the operator over the link, and anything unfamiliar hands control back to a person.
-
03
Autonomous tasks
in pilot
The machine completes defined routines with human supervision and intervention.
the gateWe validate each routine on the customer’s machine before it runs there.
-
04
Supervised fleet autonomy
in development
The goal: several machines perform validated workflows under one supervisor.
the gateEach machine runs its routine and calls for a person when anything is new, so one supervisor covers several.
Remote operation is available now. Statuses describe our programme, not results on your machine. We add an autonomous function once it clears its gate in production.
operator assistance
Begin with
repeatable movements.
Candidates include holding a tool angle, repeating a swing and positioning a load. Each function needs its own validation before use on a customer machine.
how it works
The operator picks the task. Physical AI executes the movement, inside hard safety limits that never come off.
candidate functions · in pilot
- angle hold
- One button holds the tool angle, so a hard move needs less skill.
- auto swing
- Repeat swings to the same target run themselves, at governed speed.
- auto spot
- The last metre into a hopper or truck places itself, every cycle.
- safety envelope
- The machine cannot swing into a zone you mark: walls, trucks, people.
the long-term goal
More machines working
under human supervision.
As validated workflows expand, machines can do more of the work on their own. People focus on supervision, unfamiliar situations and tasks that still require human judgement.