See, hear and sense balance
It sees in stereo, hears voices and tracks its own body movement.
Model not named
1X Technologies
How 1X NEO works
1X NEO runs Redwood AI on top of its own sensing and whole-body control — the parts the 6 public sources on this page actually name.
NEO understands you through a built-in language assistant, uses a dedicated robot-body model called Redwood to move and do chores, and separately researches a 'world model' that can imagine and evaluate outcomes.
What 1X NEO uses to sense, think, move and learn
Senses
Turns cameras, audio and touch into one picture
- Model not named
Understands
Works out what the job is
- Built-in LLM + Redwood task behaviours
Predicts
Anticipates what happens next
- 1X World Model (1XWM)
- World Model Lab
Acts
Chooses the movement and the grasp
- Redwood mobile manipulation
- Redwood AI
- Redwood whole-body
Controls
Keeps the body balanced while it works
- Redwood whole-body
Learns
Improves the models between runs
- Redwood training from EVE
- NEO data
- World Model Lab
- Expert Mode learning
What it learns feeds back into the models before the next run.
Greyed cards are parts the vendor hasn't named a model for.
01
How the robot is built to work
When it does not know how to do something, a remote human expert can step in.
Technical read
A built-in LLM handles language, memory and intent; Redwood AI is the vision-language-action policy for mobile manipulation and whole-body control; 1X World Model / World Model Lab is a distinct predictive and self-learning research direction; Expert Mode provides human tele-supervision for unfamiliar tasks.
- Redwood VLA policy
- Separate world-model direction
- Human-supervised Expert Mode
- 14
- System parts on the public record
- 13
- Parts with a public model name
- 06
- Primary architecture sources
9 confirmed as running on the robot
0 supplied by a partner
14 open questions left unanswered
NEO is one of the clearest cases where policy and world-model thinking coexist. Redwood is the deployable generalist robot model; 1X World Model / World Model Lab is the predictive/evaluation/self-learning direction, not a confirmed live runtime component. Expert Mode means some tasks can be remotely supervised by a human — that is not autonomy.
02
How it senses, reasons and moves
Three-level hierarchy
A slow reasoning layer drives a faster policy, which drives the control loop.
Onboard computer
A powerful onboard computer runs the models on the robot.
- 1X NEO Cortex (Nvidia Jetson Thor)
Physically safer body
It is built to be soft and safer if it bumps into a person.
Model not named
Run it like an appliance
The robot is operated like a home appliance with an app and a helper mode.
Model not named
Talk and remember
You can talk to NEO, and it uses memory and what it sees to understand and respond.
Model not named
Decide the task
The language brain works out what you mean, and Redwood turns it into a household task.
- Built-in LLM + Redwood task behaviours
Human remote helper
If the robot does not know how to do a chore, a human operator can step in and guide it remotely with owner permission. This is supervised human assistance, not autonomy.
- 1X Expert Mode
Choose body actions
Given what it sees and what you ask, Redwood chooses the body's actions.
- Redwood AI
Move around the home
It can move around a home as part of doing a chore.
- Redwood mobile manipulation
Grasp and handle objects
The hand story is strong but version-specific: the official product page and a newer hand article describe different hands.
Model not named
Coordinate the whole body
It coordinates walking, leaning and bracing together with using its arms, not just one at a time.
- Redwood whole-body
Physical muscles
Its muscles use tendons and cables so it can be lighter and gentler around people.
Model not named
Feeds back into the models
Around the stack
Named on the public record, but not part of the runtime chain above.
Imagine what may happen
Separately from the robot's day-to-day skill model, 1X is building a system that can imagine what a successful action should look like and use that to evaluate or learn new behaviour.
- 1X World Model (1XWM)
- World Model Lab
Learn from experience
Every robot interaction, including teleoperated help, can become future training data if captured and approved. This happens off the robot, not during live use.
- Redwood training from EVE
- NEO data
- World Model Lab
- Expert Mode learning
03
What sensors and hardware it has
Parts sit on the body only where the public record places them.
Vision and audio
02Dual 8.85MP 90Hz stereo fisheye cameras.
Two high-speed wide eyes for depth and scene understanding.
Confirmed on this robot · Head · 1 source
4 beamforming microphones; speakers in pelvis/chest.
It can hear speech and speak or play audio.
Confirmed on this robot · 1 source
Body position and balance
03Linkwise differential IMUs across the body.
It tracks its own body movement and balance.
Confirmed on this robot · Across the robot · 1 source
Degrees of freedom: hands 22x2, arms 7x2, neck 3, spine 2, legs 6x2.
A detailed body-joint breakdown is public.
Confirmed on this robot · Across the robot · 1 source
Still open: The hand article may supersede some of these joint details.
5'6" (168 cm) height; 66 lbs (30 kg) weight.
Human-sized but much lighter than most industrial humanoids.
Confirmed on this robot · Across the robot · 1 source
Hands and end effectors
01Product page: 22 DoF per hand. Separate 'NEO's Hands' article claims a breakthrough 25-DoF tendon-driven hand — treat as conflicting/evolving generation detail.
The hand story is strong but version-specific across two official sources.
Conflicting public disclosures · Hands · 2 sources
Still open: Which hand generation ships on which NEO configuration?
Joints and actuators
02Patented Tendon Drive; low-inertia, high torque-density motors with torque accuracy and backdrivability specs.
Its muscles use tendons and cables so it can be lighter and safer around people.
Confirmed on this robot · Across the robot · 1 source
Lift 154 lbs; carry 55 lbs; arm payload 18 lbs.
Strong for its weight, though this is a static spec, not proof of chore capability.
Confirmed on this robot · 1 source
Still open: Distinguish static lift specs from actual demonstrated chore capability.
Compute and connectivity
011X NEO Cortex onboard compute module (Nvidia Jetson Thor), up to 2070 FP4 TFLOPS listed.
A powerful onboard computer runs NEO's AI models.
Confirmed on this robot · 1 source
Battery and runtime
01842 Wh battery; ~4 hour runtime; quick charge adds roughly 6 minutes of runtime per minute charged.
Designed for several hours of home work before it needs to recharge.
Confirmed on this robot · 1 source
Safety hardware
02Hands rated IP68, body rated IP44; 22 dB audible noise.
Quiet operation, and hands that can handle wet tasks better than the rest of the body.
Confirmed on this robot · 1 source
Soft custom lattice polymer body; no pinch points; Head Injury Criterion (HIC) below 250.
Built to be softer and safer if it bumps into a person.
Confirmed on this robot · 1 source
Still open: Independent certification is not public.
04
How it does one real job
Vendor-stated scenario
Fetch a mug from the kitchen table and place it on the counter
This is a vendor-described scenario, not an independently verified deployment log. World-model use and Expert Mode escalation are described as possibilities, not confirmed to occur on every run.
01
Understand the request
NEO hears the request and turns it into a task goal.
Technical detail
The LLM/audio layer recognises the request and converts it into a task goal for the planner.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Built-in LLM + Redwood task behaviours
Confirmed on this robot · 1 source
02
Find the mug
It looks around the kitchen and identifies the table and the mug.
Technical detail
Visual intelligence and Redwood identify the target object and its location from stereo camera input.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Redwood AI
Confirmed on this robot · 2 sources
03
Walk to the table
It navigates across the kitchen while avoiding obstacles.
Technical detail
Redwood's mobile-manipulation policy coordinates locomotion to the table.
Home robustness across varied layouts still needs field proof.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Redwood mobile manipulation
- Redwood whole-body
Confirmed on this robot · 1 source
04
Reach and grasp gently
It reaches out and picks up the mug with compliant fingers.
Technical detail
Redwood outputs full-body manipulation actions; tendon-driven hands close under compliant control.
Exact hand generation and tactile detail should be versioned, given conflicting hand disclosures.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Redwood AI
Confirmed on this robot · 2 sources
05
Carry to the counter
It walks while keeping the mug stable.
Technical detail
Locomotion and manipulation are coordinated so the carried object stays balanced.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Redwood whole-body
Confirmed on this robot · 2 sources
06
Place the mug, or escalate to a human if stuck
It places the mug and may retain context in memory; if it cannot complete an unfamiliar step, a remote 1X Expert can take over with owner permission. Separately, 1X states the world model can be used to evaluate or improve policies, but this is not confirmed to run live on every action.
Technical detail
The robot places the mug and updates memory if enabled. Expert Mode provides tele-supervised assistance for unfamiliar tasks, distinct from autonomous execution. World-model self-learning is a separate offline/direction-level process for evaluating and improving future policies.
Do not assume every fetch action invokes the world model live, and do not count Expert Mode assistance as autonomous capability.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- 1X World Model (1XWM)
- World Model Lab
- Redwood training from EVE
- NEO data
- Expert Mode learning
Confirmed for the robot family · 3 sources
Parts are lit where the vendor's own description puts them to work. Unlit parts still run on the robot; they are simply not what this job turns on.
05
How it learns and improves
01
The robot works
Runs the parts of the system that later receive updates (3)
02
Experience is captured
Robot logs · Teleop data · Human videos · Simulation / on-policy data
03
Learn from experience
Confirmed on this robot
04
Updates go back on the robot
Model improvements · New skills
Station 04 returns to station 01 — the loop repeats.
Every robot interaction, including teleoperated help, can become future training data if captured and approved. This happens off the robot, not during live use.
Technical detail
1X describes training from teleoperated/autonomous episodes and building world-model pretraining/self-learning data loops that feed Redwood and the world-model direction.
Confirmed on this robot
- Redwood training from EVE
- NEO data
- World Model Lab
- Expert Mode learning
How user data is retained/used and opt-outs affect learning is important.
What the updates land on
- Choose body actionsActs
- Imagine what may happenPredicts
- Learn from experienceLearns
06
What the vendor has shared
- 24
- Vendor-confirmed
- 02
- Reported or contested
A map of the public record, not a verdict. A silent line means the vendor has not said — it never moves the ranking.
What each band means
Vendor-confirmed
The vendor publicly describes this for the exact robot named on this page.
Reported or contested
Public sources disagree; we show the conflict rather than pick one.
Still open · 03
Which NEO hand generation ships?
The product page states 22x2 DoF hands; a separate article describes a 25-DoF hand breakthrough. 1X has not reconciled which ships on which configuration.
Why it matters
Hand DoF and tactile capability materially affect manipulation quality.
How much is autonomous versus Expert Mode?
1X openly uses Expert Mode for unknown or complex chores; the autonomous/supervised split for any given task is not quantified.
Why it matters
Autonomy claims must be separated from human remote operation.
Is the world model live in policy execution?
1X discloses several world-model research uses (prediction, evaluation, self-learning), but the exact production runtime relationship to Redwood is evolving and not confirmed as NEO's live action path.
Why it matters
The world-model architecture is significant but should not be over-attributed to NEO's day-to-day runtime.
07
Technical details
The record
Every row is the public position. Undisclosed rows stay in the list.
08
Sources
6 primary sources sit behind this page. Where they stop, the page says so.
Vendor
The ranking measures demonstrated capability. This page explains the public system behind it. The system description does not affect the score.
Return to the 1X NEO report →