Skip to main content
Innate lets you train end-to-end manipulation policies directly from the app. You demonstrate a task with the leader arm, upload the data, train a model on Innate’s cloud GPUs, and deploy the result as a skill your robot can execute autonomously. The underlying architecture is ACT (Action Chunking with Transformers) — a neural network that observes camera images and joint positions, then outputs coordinated arm and base actions.

​
The pipeline at a glance

Record episodes  →  Upload dataset  →  Train on cloud  →  Download model  →  Run as a skill
     (app)            (~10s/min)        (1-3 hours)         (automatic)        (app or code)
Everything happens through four tabs inside a physical skill page in the app:
TabWhat you do
RecordCollect demonstration episodes with the leader arm
TrainConfigure hyperparameters and launch a training run
RunsMonitor active training jobs
CompletedDownload finished models and activate them
Once a trained model is activated, the skill appears in Manual Control and is available to agents and code.

​
What goes in, what comes out

Input: A dataset of teleoperated demonstrations — each episode captures synchronized camera images (main + wrist), joint positions, joint velocities, and optionally wheel odometry at 30 Hz. Output: A PyTorch checkpoint that runs inference at 25 Hz, outputting 6 arm joint commands and 2 base velocity commands every 40 ms.

​
When to use trained skills

Trained policies shine when:
  • The task involves visuomotor coordination (reach for, grasp, place)
  • Object positions vary between runs and the robot needs to adapt visually
  • Writing explicit motion code would be brittle or impractical
For fixed, repeatable motions (a wave, a gesture), a replay skill is simpler — record once, play back. See Policy-Defined Skills for the comparison.

​
Next steps

Record a dataset

Collect high-quality demonstrations.

Train a policy

Configure and launch training on Innate’s cloud.

Deploy your skill

Download, activate, and run your trained model.

Dataset format

Understand what’s inside each episode file.

Training Manager

Browser-based power-user UI for dataset management.