KITCHEN ROBOT NOW START A ROBOT BRIEF ↗
AI-NATIVE ROBOTICS COMPANY / 2026

SPECIALTY ROBOTSFOR REAL ROOMS

FROM IDEA TO FINALIZED CONTROLLER01
UNDER 30 DAYSdesign target / current loop

We use AI to design the body, mechanisms, physics, and controller for a real work cell—fast enough to make specialty robots radically cheaper.

FIRST PROOF / KITCHEN ROBOT NOWCURRENT FOCUS / BIOLOGY LABSAI + PHYSICS + CONTROLLERS
High-contrast concept study of a specialty robot arm CURRENT FOCUS / BIOLOGY LABS A / 01
KWR-1 kitchen work robot renderKWR-1 / FIRST PROOF01
Exploded KWR-1 assembly studyBODY / PARTS / ROUTING02
KWR-1 gripper contact studyCONTACT / FRICTION03
SPECIALTY / BY ROOM04
AI BUILDS THE BODYPHYSICS BREAKS ITCONTROLLERS LEARN
CHEAP ENOUGH TO EXIST
01 / ENVIRONMENT02 / WORK03 / BODY SEARCH
02 / REAL ROBOTS / ACTUAL EXPORTS

One loop.
Many bodies.

The kitchen robot is the proving ground. W1 Wriggler is the first robot we intend to build to test whether AI-designed bodies survive contact with reality. Biology labs are the next room already inside the design search.

KWR-1 / CURRENT PLATFORM01
KWR-1 dual-arm kitchen work robot render

Kitchen work robot.
Full stack.

Geometry, subassemblies, scene export, BOM, electronics schedules, wiring, assembly notes, and verification all leave the same pipeline.

CADBOMWIRINGSIM
OPEN KWR-1 DOSSIER ↗
W1 / WRIGGLER / FIRST BUILD TARGET02
W1 Wriggler segmented crawler render

The first build.
The sharp test.

A small worm-like crawler with printable geometry, a firmware starting point, motion videos, and 12/12 simulated gates—still explicitly unbuilt.

STLFIRMWARE12/12 SIM
OPEN W1 BUILD PACKAGE ↗
BIOLOGY LABS / NEXT ROOM03
Biology lab specialty robot concept render

New room.
Same pipeline.

Plate handling, instrument tending, sample movement, and safe repeatable interaction are the current specialty direction for the next body search.

ROOM BRIEFBODY SEARCHCONTROLLER
SEE BIOLOGY LABS ↗
THE COMPANY TESTDESIGN A BODY → BUILD ONE → SEE WHAT REALITY EDITSSTATUS / W1 UNBUILT
03 / DESIGN SEARCH / MANY BODIES

Don't pick a body.
Search the room.

We generate candidate bodies, put them into the same task suite, break them on contact and reach, then train the controller against the outcomes. The KWR-1 challenge library is the proving ground; these renders make the search legible at a glance.

KWR-1 design search visualization with ghosted body candidates in a kitchen challenge cell
AI-ASSISTED BODY SEARCH / VISUAL STUDYDERIVED FROM KWR-1 TASKS
01GENERATEbody · reach · tools
02CHALLENGEcontact · friction · failure
03TRAINobservations · outcomes · policy
04COMPILEheld-out room · controller

The body and the controller stay linked to the same environment until the design has reasons to pass—and reasons to refuse.

High-quality KWR-1 industrial design render
KWR-1 / BODY RENDERpolished visualization of the current mechanical design
High-quality KWR-1 biology lab concept render
BIOLOGY LABS / NEXT ROOMconcept visualization of the current specialty direction

GENERATED DESIGN STUDIES. THE ACTUAL TEST SURFACE IS THE CHALLENGE MATRIX, SIMULATION REPORTS, AND EXPORTED ENGINEERING PACKAGE.

02ONE ROOM
AT A TIME
THE BET / SPECIALTY HARDWARE

The next useful robot is built for one room.

Kitchen Robot Now is the first proving ground. Biology labs are the current specialty design focus: plate handling, instrument tending, sample movement, and safe repeatable interaction.

same pipeline / new room
KWR-1 biology lab specialty render
THE DESIGN LOOP / TARGET UNDER A MONTH

From environment brief to finalized controller.

TARGET30DAYS
STAGE 01 / BRIEFAI + ENGINEERING

Start with the room, not the robot.

Describe the environment, the work, and what counts as a safe failure. The same spec becomes the language for geometry, simulation, and verification.

room+work+constraintsdesign brief
The design clock starts when the room becomes legible.
THE FIRST PROOF / KITCHEN ROBOT NOW

A prototype is a chain of inspectable artifacts.

KWR-1 is the first room where the loop has teeth: geometry, contact, failure, learning, and an export package you can open.

KWR-1 challenge contact sheet32 SCENARIOS / 6 BODIES
KWR-1 gripper contact studyGRASP ONLY COUNTS WHEN FRICTION HOLDS
Expected-failure render showing an overloaded pot dropFAILURE IS AN OUTPUT
HELD-OUT / CEM0.98
VALIDATION GATE20 / 20

SIMULATED EVIDENCE, NOT A CLAIM OF PHYSICAL DEPLOYMENT READINESS.

06 / EXPORTS / RENDER → BOM → WIRING

Make the depth inspectable.

Real robotics work is not one hero render. It is the body, its subassemblies, the export formats, the wiring map, the build notes, and the reasons a gate is still open.

KWR-1 cable routing engineering renderACTUAL MODEL / CABLE ROUTING
KWR-1 / ENGINEERING EXPORT

From body render to wiring schedule.

Open the actual package: GLB bodies, subassemblies, BOM CSV, one-line wiring SVG, connector and cable schedules, assembly notes, and machine-readable verification.

OPEN THE KWR-1 PACKAGE ↗
W1 obstacle avoidance simulation plotSIMULATION / OBSTACLE AVOIDANCE
W1 / FIRST BUILD TARGET

Small enough to build. Honest enough to fail.

W1 carries the question into hardware: printable parts, firmware starting point, motion evidence, and a report that says exactly what is still only simulated.

OPEN THE W1 PACKAGE ↗

PUBLIC ENGINEERING-DEVELOPMENT SNAPSHOT / MODELED ≠ BOM-COVERED ≠ ENGINEERING-SELECTED ≠ ORDERABLE / PROCUREMENT RELEASE REMAINS BLOCKED

THE POINT OF THE LOOP

Make bespoke hardware feel more like software.

The target is not cheap as a slogan. It is a design search over body, drives, sensors, materials, and build routes until a specialty robot can exist at radically lower cost.

READ THE COST STUDY ↗
<30
OPEN NOTEBOOK / RESOURCES

Start with the artifacts.

OPEN THE VISUAL ARTBOOK ↗
BUILD THE NEXT ROOM

Have a specialty job that deserves its own robot?

Bring the environment, the constraints, and the awkward work. We are building the AI-native system that makes a robot worth testing in weeks—not years.

START A ROBOT BRIEF ↗