compareeHub

TODDLERBOT — STANFORD OPEN-SOURCE TODDLER-SIZED HUMANOID

A toddler-sized humanoid that learns tasks from human demonstrations, then repeats them — and the whole platform is MIT-licensed and printable.

by Haochen Shi, Weizhuo Wang, Shuran Song and C. Karen Liu (Stanford University)

FULL CAD BOM FIRMWARE DOCS

RoboticsAI

Built with3D printing

difficulty
●●●●●
time
weeks-plus
license
MIT
repo
repo ACTIVE780 stars

WHAT YOU’LL NEED

Videos
1
Jump to section

COMPAREE VERDICT

ToddlerBot is a toddler-sized humanoid research platform built at Stanford University by Haochen Shi and Weizhuo Wang, advised by Shuran Song and C. Karen Liu, designed for loco-manipulation tasks: robots that walk, manipulate objects, and learn from human demonstrations using diffusion policies and reinforcement learning. The January 2026 release added whole-body multi-skill locomotion. The software stack is pure Python and installs with pip, the CAD is public on Onshape, and printable files are on MakerWorld — all under an MIT licence. This is a serious research platform, not a weekend robot. The docs include a full BOM (2xc or 2xm variant), PCB files, a 3D-printing guide and an assembly manual, but the software assumes you are comfortable with Python, PyTorch and RL training in MuJoCo, and there is no hand-holding on tuning. The single thing most likely to go wrong is underestimating the integration work: even with all the files, you are building a one-off research prototype, not assembling a kit. If you are doing humanoid research or serious multi-robot manipulation experiments and need an open platform to build on, this is one of the best-documented starts available. If you want a robot that works out of the box, this is not it.

GOOD TO KNOW

  • —CAD is on Onshape (public), printable files are on MakerWorld, and the software stack is pip-installable Python.
  • —The repository and docs include training code, simulation, deployment, a full BOM, PCB files, a 3D-printing guide and an assembly manual.
  • —The structure is printable; the servos, compute and electronics are listed in the official BOM (2xc or 2xm variant).
  • —Documentation covers installation, simulation, policy training and the hardware (BOM, PCB, 3D printing, assembly manual) — but assumes significant ML and robotics background.
  • —MIT licence permits commercial use.
  • —The January 2026 release added whole-body multi-skill locomotion; this is an active research platform, not a turnkey kit.

BUILDS OF THE WEEK

Five open-source builds worth your weekend, every week.

Checked like this one: what’s really in the repo, what it costs, how hard it is. One email, unsubscribe anytime.

Can I build this?

PrintFull humanoid structure — arms, legs, torso, head — printable on a standard FDM printer; files are on MakerWorld
BuyEverything in the official BOM (servos, compute, PCB, battery, fasteners) — choose the 2xc or 2xm variant
Tools3D printer (FDM), soldering station, a computer for the Python stack (Linux, macOS or Windows; a strong GPU helps for RL training), MuJoCo
SkillsAdvanced — requires ML (diffusion policies, RL), Python, robotics kinematics, electronics integration, and the ability to debug multi-robot systems
TimeWeeks to months — printing is days, assembly is weeks, getting the software stack running and training your first policy is another week minimum
Cost$$$ — under 6,000 dollars in total according to the paper, dominated by actuators and electronics; the printed structure is cheap
SafetyActuators can pinch or crush if mis-controlled; lithium batteries if you use them for mobile power; otherwise standard electronics and mechanical assembly precautions. No mains voltage in the design shown.

Build at your own risk. Projects involve tools, electronics and sometimes mains voltage — follow the creator’s safety notes.

Videos

Primary demonstration video from the repository

More builds like this

All projects
OpenArm: A Research-Grade Force-Feedback Robot ArmRoboticsXLeRobot: A Household Robot on an IKEA CartRoboticsALOHA: The Stanford Robot That Learns From YouRobotics

Gallery

Start here

Navigation into the creator’s own docs — we don’t rewrite the guide, we route you to the source.

  1. 1.Read the repository README and watch the demonstration videos to understand what the platform actually does and what background you need (The README is technical and assumes ML + robotics familiarity — if it reads like a foreign language, this project is not a good starting point)
  2. 2.Get the official BOM (2xc or 2xm variant) and the printable files (The docs recommend ordering the 'essentials' section first; PCB files, a 3D-printing guide and the printable files on MakerWorld are linked from the docs.)
  3. 3.Install the pip package and run the MuJoCo simulation before touching hardware(The stack is pip-installable Python (3.10) with setup instructions for Linux, macOS, Windows and Jetson; replaying a keyframe motion in the MuJoCo viewer is a good first test.)
  4. 4.Print, assemble, wire, and integrate the physical robot — this is where most of the time goes(Expect troubleshooting, calibration, and multiple iterations on wiring and control tuning)

Resources

Documentation, files and community threads for this build — we link straight to the original sources and never rehost the creator’s files.

KNOWN ISSUES

  • Buying everything at once — the docs recommend ordering the 'essentials' section of the BOM first in case your plan changes.
  • The software stack assumes ML and robotics background — if you have never trained an RL policy in MuJoCo or a diffusion policy, expect a learning curve before you run your first demo.
  • This is a research platform, not a kit — there is an assembly manual in the docs, but wiring, calibration and debugging still take real time.
  • Actuators and electronics dominate the cost — the printable structure is cheap, but commercial servos and motors for a humanoid are not, and you will need several.
  • Multi-robot collaboration (two ToddlerBots working together) requires two complete builds and coordinated control — do not assume that is the starting point.
  • The January 2026 locomotion release is recent — expect the codebase to evolve, and be prepared to pull updates and adapt your build.

Is there a parts list?

Yes — the docs have a complete BOM (two variants: the recommended 2xc and the stronger 2xm), plus PCB files and an assembly manual.

Can I build this without ML experience?

You can build it: the authors say the documentation allows assembly with basic technical expertise, and an independent team replicated it. Getting value from it, though — training RL or diffusion policies — assumes you are comfortable with Python and PyTorch.

How much does it cost to build?

The paper puts the total cost under 6,000 dollars. The printed structure is cheap; the servos, onboard computer and electronics in the official BOM make up most of it.

Does it work out of the box?

No. This is a research platform. You print, source parts, assemble, wire, calibrate, and train policies yourself. There is no turnkey assembly.

Can it really learn tasks from human demonstrations?

Yes — that is the point of the platform. You collect human demonstrations, train a diffusion policy, and deploy it to the robot. But training, sim-to-real transfer, and tuning are all part of the work.

What is the locomotion update?

The January 2026 'Locomotion Beyond Feet' release adds a multi-skill whole-body locomotion system: the robot uses stereo depth and a skill classifier to pick the right trained skill and get over different obstacles on its own.

Community builds

No community builds yet — be the first, we feature the best ones.

Discussion1

FROM THE COMPAREE TEAM

A full BOM, an assembly manual, Onshape CAD and MakerWorld prints — plus RL training code that runs from pip. If you built one, would you start with the walking policies or the manipulation side?

CompareeTEAM2mo agoedited

Practical notes from our verification: the CAD is public on Onshape, the printable files are on MakerWorld, and the code is pure Python and fully pip-installable — and, contrary to what you might expect from a research robot, the documentation does cover the build: there is a bill of materials (with the recommended toddlerbot_2xc and the toddlerbot_2xm variants), PCB notes, a 3D-printing guide and an assembly manual, linked from the docs site. The README itself is aimed at people training policies — reinforcement learning in MuJoCo, diffusion-policy training and real-world deployment — so expect research tooling rather than a hobby-kit tone. The codebase is still evolving: ToddlerBot 2.0 landed in August 2025 and the Locomotion Beyond Feet multi-skill locomotion release in January 2026. If you are doing humanoid research and need an open platform, this is a strong start; if you want a weekend build, the parts list and assembly manual are there, but budget real time for printing and wiring. Correction (4 October 2026): we re-checked this page line by line against the project's own repository, documentation and videos, and fixed errors in earlier versions.

Haochen Shi, Weizhuo Wang, Shuran Song and C. Karen Liu (Stanford University)

ToddlerBot was built at Stanford by Haochen Shi and Weizhuo Wang, advised by Shuran Song and C. Karen Liu, as a low-cost humanoid for ML research — and released open-source under MIT so other labs can build on it.

GitHub

Star the project on GitHub

DISCLAIMER

  • Comparee is not the author of the projects featured here. All rights to each project belong to its creator — every page links to the original source, and we never host creators’ files.
  • Information is provided without warranty and may become outdated as projects evolve. Prices are indicative bands only — always check the creator’s parts list for current costs.
  • Building and operating any project is at your own responsibility. Protective equipment, safe workshop practice and compliance with local regulations are the builder’s responsibility.
GitHub repo