Human skill. Machine ability.

Anyone can
teach a robot.

The next era of robotics:
skills are learned, not programmed.

One learning system to capture human demonstrations, train adaptable behaviours, and bring them to physical machines.

Open Imitation MachinesRobot Learning StudioSee it in practice
Teach a robot by showing itA line drawing of a mobile robot arm. An Imitation Machines module is installed on it and its cameras stream into the Imitation Machines interface. A person demonstrates picking up a machined part and placing it in a bin with a VR controller while the arm follows and each demonstration is recorded. The platform trains on the demonstrations, Deploy is clicked, and the robot performs the task on its own.ZXREACHθ1θ2IM MODULECompute, sensing, linkE-STOPWRIST CAMERALIDARMOBILE ROBOTS SUPPORTEDVR CONTROLLER6-DoF trackedOPERATORAUTONOMOUSLearned policy, no operatorROBOT LEARNING STUDIORobot 01 · mobile arm(offline)VR controllerConnectedAutonomous modeLIVELIVEPick up the part and place it in the binPick and placeDataset201 episodes · 1h 08m · new episodes are saved hereTraining · ACT0%Learning from 3 new demonstrationsDeploy to robot201 / 10201 episodes · 68.4 min#202 Oct 23, 10:41 AM ( Duration: 0:18)#203 Oct 23, 10:42 AM ( Duration: 0:16)#204 Oct 23, 10:42 AM ( Duration: 0:17)#201 Oct 8, 02:43 PM ( Duration: 0:18)#200 Sep 28, 07:11 PM ( Duration: 0:34)#199 Sep 28, 07:11 PM ( Duration: 0:18)#198 Sep 28, 07:10 PM ( Duration: 0:18)#197 Sep 28, 07:10 PM ( Duration: 0:22)#196 Sep 28, 07:10 PM ( Duration: 0:17)#195 Sep 28, 07:06 PM ( Duration: 0:17)#194 Sep 28, 07:06 PM ( Duration: 0:21)#193 Sep 28, 07:05 PM ( Duration: 0:18)#192 Sep 28, 07:06 PM ( Duration: 0:16)#191 Sep 28, 07:05 PM ( Duration: 0:19)#190 Sep 28, 07:05 PM ( Duration: 0:19)TRAINING POLICYEPOCH 0 / 40DEMONSTRATIONSLOSSProcessing 204 episodes…0%
How it works, in four steps.From human demonstration to machine action.

The real world

Reality doesn’t
repeat itself.
Robots need to adapt.

An object shifts. A material bends. The next task starts a little differently. Real work is full of variation.

We’re building a learning system that captures how people respond to that variation, so machines can learn more than a fixed sequence of movements.

Variation is the default.Demonstration carries context.Experience improves behaviour.

The learning loop

Show. Learn. Deploy. Improve.

A repeatable path from knowing how
to doing it on a machine.

01

Show.

Start with the person who knows the work. Capture their demonstrations and the context around them.

02

Learn.

Turn examples into behaviour. Train a policy, evaluate it, and focus on what needs to improve.

03

Deploy.

Bring the learned skill to the machine. Observe how it performs in the conditions that matter.

04

Improve.

Bring interventions and new edge cases back into training. Keep building on what was learned.

Demonstration → behaviour → field experience → demonstration

A wide horizon

One approach.
Unbounded application.

For work that changes with the objects, the environment, and the day.

  1. 01

    Manufacturing

    Assembly & material handling

  2. 02

    Logistics

    Picking & movement

  3. 03

    Agriculture

    Natural variation

  4. 04

    Construction & maintenance

    Changing environments

  5. 05

    Labs & research

    Careful, repeatable handling

  6. 06

    Homes & everyday help

    Tasks that differ every day

  7. 07

    And beyond

    Any robot, any task

The next behaviour starts with you

What could you
teach a machine?

Bring a task, a robot, or a challenge.
Let’s explore what learning can unlock.

Tell us about your taskhi@imitationmachines.com

Working in partnership with

  • The University of Sydney
  • NVIDIA Inception Program
  • The University of Newcastle
  • Generative Ventures
  • FaBA, Food and Beverage Accelerator
  • Undisclosed industry partner
  • Antler
  • Undisclosed industry partner
  • Atlas SGR
  • Undisclosed industry partner
  • The University of Queensland

From our workbench

Real machines.
Learning new moves.

Early experiments. Tangible progress. A look at learning across different tasks and hardware.

Every demonstration here was trained on our platform, without writing a line of code, using only the Robot Learning Studio.

Physical trial / Pick and place01

Placing garlic into a basket on a conveyor.

reBot B601-RS arm. Policy trained with Diffusion Policy. The wrist camera, shown in the inset, is its only visual input.

Physical trial / Bimanual mobile02

Putting a plate away with two arms.

Two arms on a mobile cart lift a plate from the dining table and put it away in a kitchen drawer.

Simulation / Dexterous grasp03

Moving a kettle with a five-fingered hand.

UR10e arm with an Inspire RH56 hand in MuJoCo. Policy trained with ACT on demonstrations recorded with a SenseGlove.

Physical trial / Sorting04

Sorting wooden shapes onto a puzzle board.

SO-100 arm. Each shape goes into its matching slot. Learned from 200 demonstrations, collected in 1 h 49 min.

Prototype research footage. These demonstrations illustrate individual experiments, rather than production performance.