Robotiqs

Get a working robot policy in 5 days — without the GPU hell.

You send your SO-101 teleoperation demos. We train and validate a real ACT policy on cloud GPUs and send it back ready to run. No Docker, no CUDA, no $30k workstation.

Fixed price · Delivered in 5 business days · Refunded if it doesn't work

Training a robot policy is mostly setup, not robotics

You bought an arm and recorded demonstrations. Turning those into a working policy means installing CUDA, fighting Docker, renting and configuring a GPU, and learning a command-line training pipeline. Most makers and small teams stall right here — the robot never moves.

Robotiqs removes that entire layer. You stay in the part you care about — your robot and your task. We handle the training.

How it works

1

Send your demos

Upload your LeRobot-format dataset of teleoperation episodes — 50–200 demos for a typical task.

2

We train & validate

We train an ACT policy on cloud GPUs and validate it on a physical SO-101 — not just in simulation.

3

You get a ready policy

You receive the trained policy plus run instructions, ready to deploy on your robot — within 5 business days.

Simple, fixed pricing

No subscription. Pay per policy. Refunded if we can't deliver a working one.

Starter

$199

One task, one trained policy

  • ✓ One ACT policy trained on your data
  • ✓ Validated on physical hardware
  • ✓ Run instructions included
  • ✓ Delivered in 5 business days

Team

$499

Multiple tasks or an iteration cycle

  • ✓ Multiple policies, or iteration on one task
  • ✓ Validated on physical hardware
  • ✓ Run instructions + tuning notes
  • ✓ Priority — delivered in 5 business days

Why Robotiqs

Built on LeRobot. We use the open-source standard for robot learning — your dataset and the policy you get back are standard formats, not locked to us.

Validated on real hardware. Every policy is run on a physical SO-101 before we send it — not signed off in simulation.

Refund guarantee.If we can't produce a working policy from your data, you don't pay.

Questions

Do I need a GPU?

No — that is the whole point. Training runs on cloud GPUs we manage. You just need your robot and your demonstrations.

What do I send you?

A LeRobot-format dataset of teleoperation demonstrations. For a simple task, 50–200 episodes is typical.

Which robots do you support?

The SO-101 / SO-100 arm to start. Working with a different arm? Email us first and we will tell you if we can help.

What if the policy does not work?

We validate every policy on a physical robot before delivery. If we cannot produce a working policy from your data, you are refunded.

How long does it take?

Within 5 business days of receiving your dataset.

Stop fighting the toolchain. Get your robot moving.

Send your demos today — get a working policy back this week.