1X NEO teleoperation: how Expert Mode works, and what it means for privacy

The 1X NEO is a 5 ft 6 in, 66 lb humanoid built for the home, and the most interesting thing about it is not the hardware. It is that 1X built remote human operation into the product and named it: Expert Mode. When NEO meets a chore it does not know, the owner can schedule a 1X Expert to drive the robot through it remotely, and each of those sessions becomes training data for the next time. That is the same human-in-the-loop model most humanoid programs run behind closed doors, offered openly to consumers. It raises real questions about latency, labor and privacy, and this page takes them one at a time.

Can you buy a 1X NEO yet?

You can preorder one. The 1X order page lists two options as of October 2026: Early Access ownership at $20,000 with a three-year warranty, or a $499 monthly subscription, both reserved with a fully refundable $200 deposit. The page says "US Deliveries start 2026". The official FAQ adds that deliveries go to Early Access buyers first, then subscribers, with international deliveries later.

In its April 30, 2026 factory update, 1X said its Hayward, California plant can build up to 10,000 NEOs a year, that the October 2025 launch "booked out our entire production capacity for the next year in just 5 days", and that current output was going to its internal team for development and home testing before customers. We could not find independent confirmation that customer units had shipped as of October 2026. If you are planning around a delivery date, plan around 1X's written commitments, not demo videos.

How NEO is driven: autonomous by default, human on request

NEO has three ways of being told what to do, and only one of them involves a remote person.

There is no public SDK, so owners cannot attach their own teleop stack or pull raw episodes off the robot. NEO is a closed consumer product, and Expert Mode is the only remote-operation path.

Expert Mode, step by step

Putting together the official FAQ on the NEO product page and Road to VR's October 2025 report, a session works like this:

Human-in-the-loop as a business model

Expert Mode turns what is usually a hidden cost into a product feature. A humanoid that is not yet fully autonomous still has to be useful to the person paying for it, and the honest way to bridge the gap is to have a skilled person finish the job through the robot. 1X's design makes three bets that apply to anyone building a humanoid business:

Wikipedia's 1X Technologies entry records that the Wall Street Journal's Joanna Stern reported in October 2025 that most tasks she saw were teleoperated by a person in a VR headset, and that CEO Bernt Børnich said early-adopter data is needed to improve autonomy. Read plainly, that is the business model working as designed at an early stage, not a hidden trick, and it is the same arc our human-in-the-loop guide for humanoids describes for industrial fleets.

Privacy: what 1X says, and what to ask

A robot with stereo cameras, four microphones and a remote operator in your kitchen is a fair thing to be careful about. Here is what 1X commits to in its FAQ, stated as it states it:

Road to VR's report adds the trade-off in Børnich's own words: "If we don't have your data, we can't make the product better." The article framed privacy as a sliding scale, where more shared data means a more capable NEO. That is a fair description of how every learning robot works, not just this one. The questions worth asking before you invite any remote-operated robot into a home or business are the same: who can connect, when, from where, what they see, what is recorded, how long it is kept, and how to turn it off. 1X has published answers to most of those; retention periods for Expert sessions are the main one we could not find on its pages.

What remote operation over home internet has to handle

1X has not published how its Expert link is built, so this is the general picture rather than NEO's internals. A remote operator in a headset is driving through a home connection or 5G, so the link has more jitter than a lab network. The workable pattern on any humanoid is a split: the robot's own controller keeps balance and smooths commands locally, and the remote person directs reaching, grasping and where to walk. Household chores (picking up, wiping, folding, carrying) tolerate a few hundred milliseconds far better than fine assembly does. 1X keeping its Experts in the US also keeps them on a domestic round trip to US homes. Our teleoperation latency guide gives the budgets by task.

Public data from 1X

1X publishes the world_model_tokenized_data set on Hugging Face under Apache-2.0: about 100 hours in its v2.0 train and validation splits, with video stored as NVIDIA Cosmos tokens at 30 Hz plus joint states, hand open or closed state and body velocities, built for 1X's world model compression challenge. The card does not name the robot, and it is not raw teleop episodes, but it is a commercially usable view of how 1X structures humanoid data.

NEO against the humanoids you can deploy now

NEO is a finished consumer product: closed software, a managed remote-operator service and a home focus. Teams that want to run their own teleoperation, own the episodes and train their own policies need an open research humanoid instead, such as the Unitree G1, which ships with an SDK on its EDU version and an open XR teleop stack. The two answer different questions. NEO shows what the operator layer looks like when a maker sells it; a research humanoid is where you build that layer for your own product. For the home market more broadly, see our home and hospitality page.

How RoboPilots fits

We supply the Expert layer to humanoid makers and robot-learning teams, on their own hardware, without asking them to build a remote workforce. Several humanoid makers have sent RoboPilots robots to test our operating protocols. For each program we train a dedicated group of operators on the customer's unit and task protocol, and the same group stays on it.