Frequently asked questions

The questions people actually ask before and after signing up, answered concretely. If your question is missing, the contact form reaches a human, and the rest of the documentation covers each topic here in more depth.

Last updated 2026-08-09

Quick facts

The short answers first. Everything in this table is expanded in a full question below or on a dedicated documentation page.

TopicShort answer
Client pricing15.00 EUR per hour pay-as-you-go, or Starter at 99.00 EUR per month including 10 hours
Operator payout minimum10.00 EUR via Stripe Connect, typically 2 to 7 business days
Supported robotsSO-100, Koch v1.1, Franka FR3, FP3 and Panda, WidowX-250, ALOHA ViperX-300
Dataset formatLeRobot: Parquet, per-camera videos, and JSON metadata
Data ownershipCollected demonstration data belongs to the client
RolesCLIENT, OPERATOR, ADMIN
Control latency30 to 70 ms end to end

Questions and answers

What does AY-Robots cost?

Clients pay 15.00 EUR per hour of teleoperation on pay-as-you-go, or 99.00 EUR per month on the Starter plan, which includes 10 hours. Session time is tracked to the minute and billed on a monthly invoice via Stripe. Operators pay nothing; they earn at the hourly rate they set themselves.

Do I need my own robot hardware?

As a client, yes: you register your own robots on the platform with their control and stream URLs. As an operator, no: you need a computer and a good internet connection, and you control robots that clients have registered. To just try teleoperation, the /live demo works without owning anything.

Which robots are supported?

Currently the SO-100 / SO-ARM100, Koch v1.1, Franka Research 3, Franka Production 3, Franka Emika Panda, Interbotix WidowX-250, and the ALOHA ViperX-300 6DOF. If your arm is missing, submit a robot request; the supported robots page has details on each model.

How do I become an operator?

Sign up, choose the operator role during onboarding, and complete your profile with country, timezone, and your hourly rate, plus the built-in speed test. Then request certification for the robot types you want to operate; an admin reviews each request. Once certified, you can start sessions on any available robot of that type.

How fast is the control loop?

End-to-end latency is 30 to 70 ms, which is fast enough for deliberate manipulation work. Your own connection matters: for operating we recommend at least 50 Mbps down, 20 Mbps up, and under 50 ms latency to the nearest server, all of which the built-in speed test checks.

What happens if the operator goes inactive or the connection drops?

The platform monitors operator input during active sessions. After a period of inactivity a warning is sent, and if input does not resume, the session is stopped automatically. This prevents the client from being charged for idle time and does not leave the robot under a stale control lease.

Who owns the collected data?

The client. Every demonstration recorded in your sessions belongs to you, including commercial use in AI training. Operators are paid for their time and hold no rights to the datasets they helped record.

What format are the datasets in?

The LeRobot format: Parquet files for joint and action streams, one video per camera per episode, and JSON metadata (info.json, episodes.jsonl, tasks.jsonl). Joint values are normalized to -100 to 100 as a percentage of the calibrated range, not degrees. This loads directly into LeRobot-based training code.

Can I try the platform without a robot?

Yes. The /live demo lets you drive a robot arm from the browser without registering hardware or committing to anything. It is the fastest way to get a feel for what teleoperation on the platform is like.

How do operator payouts work?

Earnings from completed sessions accumulate in your balance at your hourly rate. You withdraw from Dashboard > Earnings once the balance reaches at least 10.00 EUR. Payouts run through Stripe Connect, which requires a one-time onboarding with bank details and identity verification, and typically arrive within 2 to 7 business days.

Can I train policies without my own GPU?

Yes. Training runs on cloud GPUs managed by the platform: pick a dataset, pick a policy type (ACT, Diffusion Policy, SmolVLA, or a GR00T fine-tune), and start the run. Progress, logs, and checkpoints live on a dedicated run page, and the finished policy can be run on your arm from the cockpit.

How many episodes do I need before training is worth it?

As a rule of thumb for one tabletop task: around 50 episodes gets you a policy that works from starting states close to your demonstrations, and 100 to 200 episodes gets usable robustness across the workspace. Below that, treat runs as pipeline tests rather than attempts at a working policy.

Is the platform suitable for research?

Yes, and it is one of the main use cases. Datasets are standard LeRobot, so they work with the open-source ecosystem; episodes carry Success, Recovery, and Failure labels for curation; merging combines recordings across sessions and arms; and everything is scriptable through the REST API.

How do refunds and disputes work?

If a session had real problems, open a dispute for it in the dashboard and describe what went wrong. An admin reviews the session events and data and decides on a full or partial refund, which always goes back to the original payment method via Stripe.

Is there an API?

Yes, a REST API under /api with the same capabilities as the dashboard. Authenticate with an API key (prefix ayr_live_) from /dashboard/settings, or use the hosted MCP server at https://www.ay-robots.com/api/mcp if you are wiring up an LLM agent. The API reference documents every endpoint.

How is my data protected?

All traffic is TLS-encrypted, including video and control streams. Access is role-based and enforced server-side, database rows are protected with row-level security, and session activity is written to an audit trail. Payment data is handled exclusively by Stripe and never stored on platform servers.