Robot form follows function

A research concept: generating robots that fit a factory's floor, task and constraints — and proving they work in simulation before any hardware exists.

Template Robotics is an independent research project and public design document by Dmitrii Gusev, first published May 4, 2026. It is not a commercial offering and is not affiliated with any employer, past or present.

First published: May 4, 2026 · Last updated: August 3, 2026

View the Demo
STEP 1INPUTSTEP 2WORLD & ROBOT BUILDINGSTEP 3DATA COLLECTION & TRAININGSTEP 4VALIDATIONINPUT #1Process DocsSpecs, order, tolerances& safetyINPUT #2Product PromptInsider knowledge notcaptured anywhere elseINPUT #3iPhone Capture4K60 + LiDARGenerated WorldsLife-like visuals · mm-scale precisionSimulated HardwareSim-BackedData Collection90%+ cheaper than operatorsAI TrainingBring your policy or use existing onesConceptDesigns and policies validatedin simulation beforehardware exists
▸ The Problem

Hardware teams demand factory's flexibility while existing physical AI solutions are limited by data availability.

SIMULATION
Robotics teams gave up on simulation
Simulated visuals are game-like and do not match real-world demands.
DATA
Teams don't collect data where it is born
Factory already has the data — process docs, real environments, real products.
COST
Data gap costs industry too much
U$48/hr/pilot, 10 man-hours just to validate one task-specific AI model. World models and generalist policies are years and millions $ away from bridging data gap.
▸ The Approach

The research direction: close the sim-to-real gap by bridging real-to-sim first.

The concept: derive purpose-built, simulation-validated automation designs from process documentation and environment capture. This page describes the research to date — a browser-first simulation prototype and environment-capture demo — and the longer-term concept around them.

SIMULATIONResearch prototype (open source)
Construct

State-of-the-art simulation environment running directly in your browser. Policies trained in simulation see what real robots see.

▸Open-Source
▸Web-browser first (even on iPad)
▸Bring your custom robots or build one
▸Deploy software directly in containerized workflows
▸Life-like visuals backed by the latest advancements in computer graphics
▸Sensor-level realism
IPHONE APPEarly prototype
Capture environments with mm-scale and visual precision

Guided workflow for capturing your factory floor. Policies trained in simulation see what real robots see.

▸Guided workflow
▸4K60 captures
▸ARKit support
▸LiDAR for physical world grounding
▸Camera calibration
AI-NATIVEConcept
AI-Native Workflows

As easy as your Claude Code session. Determine what data collection really matters.

▸Process engineering documentation parser
▸Monitor robot's behavior using a custom MCP
▸Foxglove integration
▸Supports cloud and local providers
HARDWAREConcept
Component Library

Concept: a library of accurately simulated standard components for composing and validating automation designs in simulation.

▸Fully supported in the simulator
▸Bring your own components via environment capture
▸ About the author

Background: factories in the US and China; purpose-built automation and humanoids.

Tesla
▸Material flow
▸Robot integration
▸4680 battery line
▸Optimus
Industrial Next · YC W22
▸Custom camera hardware
▸Autonomous robotic deployments across North America and China
Sanctuary AI
▸Next-generation humanoids
Open Source
▸Built on open-source MuJoCo with open-source contributions to the rendering pipeline
Author

Dmitrii Gusev has spent his career shipping robotics and manufacturing systems at Tesla, Industrial Next (YC W22), and Sanctuary AI. He's brought up factory lines at Fremont, Nevada, and GigaTexas, built electrical systems for Optimus, deployed autonomous robotic solutions in North America and China, and led software for next-generation humanoid robots. He started the Template Robotics research project after seeing firsthand why the factory bringup problem — not the robot — is what's holding automation back.

"After working directly with Chinese manufacturing teams, I realized why they move fast — their hardware teams work directly with vendors spinning up new factory lines and iterating new designs at scale on demand. In North America, that process takes months and millions. I started this research to close that gap — not with a humanoid robot, but with simulation-first purpose-built automation that makes factory bringup as fast as the iteration speed demands."
▸ Vision

Inspired by science-fiction. Validated through real-world experience.

Today, this research designs and validates robotic deployments in simulation. The long-term vision: manufacturers of any size create new automated production lines on demand — no complex humanoids, no months-long bringup, no factory retooling.

Imagine a future where any manufacturer can meet domestic demand from hardware teams by upgrading their production lines with physical AI — without altering existing processes. Robot's form must follow function to be truly efficient. New systems will be generated from a predefined set of modular and validated components to meet the demand of manufacturers of any scale at any location.

System effectiveness, machine learning and controls are validated in simulation before anything is built.

View the Demo