See and feel
It uses eyes in the head, eyes in the hands, touch in the fingers and a sense of its own body position.
Model not named
Figure AI
How Figure 03 works
Figure 03 runs Helix System 2 and Helix System 1 on top of its own sensing and whole-body control — the parts the 3 public sources on this page actually name.
Figure uses a three-layer system: one layer understands the job, one turns everything the robot sees and feels into movement, and one handles fast balance and actuator control.
What Figure 03 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
- Helix System 2
Predicts
Not disclosed
- Not disclosed
Acts
Chooses the movement and the grasp
- Helix System 1
- Helix locomotion within the full-body policy
Controls
Keeps the body balanced while it works
- Helix System 0
Learns
Improves the models between runs
- Model not named
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
Figure uses a three-layer system: one layer understands the job, one turns everything the robot sees and feels into movement, and one handles fast balance and actuator control.
Technical read
Helix System 2 provides semantic reasoning, System 1 produces full-body joint targets at 200 Hz, and System 0 executes at 1 kHz.
- Integrated stack
- Whole-body policy
- First-party models
- 11
- System parts on the public record
- 08
- Parts with a public model name
- 03
- Primary architecture sources
6 confirmed as running on the robot
0 supplied by a partner
13 open questions left unanswered
Figure has one of the cleanest public layer disclosures: Helix 02 maps directly onto a reasoner, an action policy and a whole-body controller. It does not publicly centre a separate world model.
02
How it senses, reasons and moves
Three-level hierarchy
A slow reasoning layer drives a faster policy, which drives the control loop.
Stay within limits
The body is being made safer and easier to live with.
Model not named
Understand the job
This is the part that understands the instruction and the situation, and decides the next sensible step.
- Helix System 2
Choose movement
This is the fast action brain: what should the arms, legs, torso and hands do right now?
- Helix System 1
Handle objects by touch
The robot can use hand cameras and touch when its head cannot see the object properly.
Model not named
Navigate and use skills
It does not only stand and pick — it walks around the room while it works.
- Helix locomotion within the full-body policy
Control the body
This is the balance and reflex layer that keeps movement smooth and stable.
- Helix System 0
Run the models
The reviewed official pages do not say what compute is on board.
Not disclosed
Predict what may happen
No separate 'imagine possible futures' model is publicly named for this robot.
Not disclosed
Physical body
The models connect all the way down to the body itself.
Model not named
Feeds back into the models
Around the stack
Named on the public record, but not part of the runtime chain above.
Improve the models
The robot's models learn from human movement and from simulation before real use. This happens off the robot, not while it is working.
Model not named
03
What sensors and hardware it has
Parts sit on the body only where the public record places them. 1 of 9 entries are still not publicly disclosed.
Vision and audio
02Redesigned camera architecture: twice the frame rate, a quarter of the latency and 60% wider field of view per camera versus the previous generation.
Better eyes give the action model faster and wider vision.
Confirmed on this robot · Head · 1 source
Improved audio system for voice interaction and reasoning.
Better microphones and speakers for talking to the robot.
Confirmed on this robot · 1 source
Touch and force
01Fingertip tactile sensing feeds the action policy directly.
Touch is an input to the movement brain, not a separate add-on.
Confirmed on this robot · Hands · 1 source
Body position and balance
01Full-body proprioception feeds System 1 and System 0.
The robot knows where its own limbs are while it works.
Confirmed on this robot · Across the robot · 1 source
Hands and end effectors
01Palm cameras and custom fingertip tactile sensors; touch sensitivity as small as 3 g claimed.
The hands can see close up and feel very small contact forces.
Confirmed on this robot · Hands · 2 sources
Joints and actuators
01System 1 outputs full-body joint targets; System 0 tracks them at 1 kHz.
The fast brain sends body targets and the reflex brain executes them smoothly.
Confirmed on this robot · Across the robot · 1 source
Compute and connectivity
01Not disclosed on the reviewed Figure 03 and Helix pages.
We do not know the onboard compute from official material.
Not publicly disclosed · 2 sources
Battery and runtime
01UN38.3-certified battery with safety layers; wireless inductive charging.
A safer battery pack, and charging without plugging in manually.
Confirmed on this robot · 1 source
Still open: Runtime and capacity are not stated.
Safety hardware
01Soft goods, multi-density foam coverings and home-oriented materials.
A softer exterior for working close to people at home.
Confirmed on this robot · 1 source
04
How it does one real job
Controlled demonstration
Unload a dishwasher and place bowls and cups into cabinets
This task has been shown in a controlled setting. That is different from repeated production deployment.
01
Understand the instruction
The robot interprets the request and the kitchen scene.
Technical detail
System 2 interprets the language instruction together with the scene.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 2
Confirmed on this robot · 1 source
02
Sequence the subtasks
It breaks the job into an ordered set of steps.
Technical detail
System 2 emits semantic goals — walk to the dishwasher, open the rack, pick a bowl, carry it to the cabinet.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 2
Confirmed on this robot · 1 source
03
Walk and reach together
It moves across the kitchen while using its arms.
Technical detail
System 1 outputs full-body joint targets at 200 Hz; System 0 executes balance and contact at 1 kHz.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 1
- Helix locomotion within the full-body policy
- Helix System 0
Confirmed on this robot · 1 source
04
Handle occluded or fragile items
It grasps items the head cameras cannot see clearly.
Technical detail
Palm-level vision and fingertip touch regulate the grasp when the head view is blocked.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
No named model for this moment.
Confirmed on this robot · 2 sources
05
Carry while balancing
It keeps the object stable while walking.
Technical detail
The full-body policy and controller adjust legs, torso and arms as the carried object changes the balance.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 1
- Helix System 0
Confirmed on this robot · 1 source
06
Place into the cabinet
It puts the item in the right place.
Technical detail
System 2 specifies the destination; System 1 and System 0 execute the placement.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 2
- Helix System 1
Confirmed on this robot · 1 source
07
Continue through the whole task
It runs the sequence continuously rather than performing one isolated action.
Technical detail
The public demonstration describes a four-minute continuous task with no resets or human intervention.
External and deployment repeatability evidence is still needed.
- Senses
- Understands
- Predicts
- Acts
- Controls
- Learns
- Helix System 2
- Helix System 1
- Helix System 0
Controlled demonstration · 1 source
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 (2)
02
Experience is captured
Human motion · Simulation · Robot fleet data
03
Improve the models
Confirmed on this robot
04
Updates go back on the robot
Updated System 0 and System 1 policies
Station 04 returns to station 01 — the loop repeats.
The robot's models learn from human movement and from simulation before real use. This happens off the robot, not while it is working.
Technical detail
Helix System 0 uses retargeted human motion plus simulation and reinforcement learning; Figure positions Figure 03 as built to learn from humans, with BotQ as the manufacturing and fleet direction.
Confirmed on this robot
Data scale and model update cadence are not public.
What the updates land on
- Choose movementActs
- Control the bodyControls
06
What the vendor has shared
- 17
- Vendor-confirmed
- 03
- The record is silent
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.
The record is silent
The public record does not say. We leave it visible as unknown.
Still open · 03
Where is a separate world model?
Figure has not publicly disclosed a standalone world model.
Why it matters
Helix is a strong stack, but it is not the same thing as a JEPA, Cosmos or 1X-style world model.
What is the field deployment proof?
Figure shows strong demonstrations; customer and home field evidence is still emerging.
Why it matters
A controlled demonstration is not the same as production reliability.
What are the complete hardware specifications?
Official pages describe sensory and control advances but not every specification.
Why it matters
Readers will ask for dimensions, battery, compute and payload.
07
Technical details
The record
Every row is the public position. Undisclosed rows stay in the list.
08
Sources
3 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 Figure 03 report →