You don’t need a $100K budget to work with humanoid robots. The open-source community has produced remarkable projects ranging from affordable full builds to sophisticated software frameworks. Here are seven worth your time.
Hardware Projects
1. InMoov — The Original Open-Source Humanoid
What: Full-size humanoid robot, 3D-printable Cost: ~$1,500-3,000 in parts License: CC-BY-NC
The granddaddy of open-source humanoids. InMoov has been around since 2012 and has the largest community of any open-source humanoid project. Every part is 3D-printable, and the build documentation is extensive.
Best for: Makers and educators who want a full-size humanoid without industrial-grade requirements. The community forum has thousands of builders sharing modifications and solutions.
Limitation: Not designed for autonomous operation or AI-driven control. Primarily a teleoperated/scripted platform.
2. Stompy by K-Scale Labs
What: Open-source bipedal humanoid designed for AI research Cost: ~$10,000-15,000 in parts License: MIT
A serious research platform. Stompy is designed from the ground up for VLA model deployment and reinforcement learning. Includes custom actuators, integrated compute (Jetson Orin), and a full simulation model in MuJoCo.
Best for: Research labs wanting a lower-cost alternative to Unitree G1 with full hardware customization. The simulation model means you can start software development before building the hardware.
3. Koch v1.1 — Open-Source Robot Arm
What: Low-cost 6-DOF robot arm for manipulation research Cost: ~$500 in parts License: Apache 2.0
Not a humanoid, but essential for manipulation research. Koch is the standard open-source arm used with LeRobot for data collection and policy deployment. If you’re starting in robot learning, this is the most accessible hardware entry point.
Best for: Anyone starting in robot manipulation. Build one in a weekend, start collecting demonstration data by Monday.
Software Projects
4. LeRobot (Hugging Face)
What: End-to-end robot learning framework License: Apache 2.0 GitHub Stars: 15K+
Already covered in our tutorial, but it deserves a spot on this list. LeRobot is the most important open-source project in robot learning right now. Data collection, policy training, model sharing — all in one framework.
Contribute: The project actively welcomes contributions. High-impact areas include new policy architectures, hardware driver support, and benchmark tasks.
5. OpenVLA
What: Open-source Vision-Language-Action foundation model License: MIT Models available: 7B, 3B, 1B variants
The community’s answer to proprietary VLA models like GR00T and pi0. Trained on the Open X-Embodiment dataset, OpenVLA provides pre-trained weights, training code, and fine-tuning scripts.
Contribute: Data quality improvement (the OXE dataset has inconsistencies), new model architectures, and benchmark evaluations are all active areas.
6. MuJoCo Menagerie
What: Collection of high-quality robot simulation models License: Apache 2.0
A curated set of robot models for MuJoCo, including humanoid robots (Unitree G1, H1), robot arms (UR5, Franka), and hands (Shadow, Allegro). These models are research-grade — accurate dynamics, proper collision geometry, and calibrated actuator models.
Best for: Anyone doing simulation-based robot learning. Instead of building your own robot model (error-prone and time-consuming), start with a validated model from the Menagerie.
7. Isaac Lab
What: GPU-accelerated robot learning framework License: BSD-3 By: NVIDIA
The open-source layer on top of Isaac Sim. Provides pre-built RL training environments, domain randomization pipelines, and integration with popular RL libraries. While Isaac Sim itself requires an NVIDIA GPU, Isaac Lab’s abstractions make it the most productive environment for large-scale robot policy training.
Contribute: Custom environments, new task definitions, and reward function engineering. The framework is designed for extensibility.
How to Choose
| If you want to… | Start with |
|---|---|
| Build a physical robot | InMoov (fun) or Koch (research) |
| Train robot AI policies | LeRobot + MuJoCo Menagerie |
| Work on VLA models | OpenVLA |
| Train at scale | Isaac Lab |
| Do everything | Koch + LeRobot + MuJoCo → Isaac Lab for scaling |
Contributing to Open-Source Robotics
The embodied AI open-source community is smaller and more impactful per contributor than web/mobile open source. High-value contributions:
- Bug reports with reproduction steps — Especially for sim-to-real discrepancies
- New simulation environments — Task diversity is a bottleneck
- Documentation and tutorials — The field grows faster when onboarding is easier
- Benchmark results — Reproducible comparisons across methods
- Hardware driver support — Making LeRobot work with new robot platforms
Every major robotics company monitors these open-source projects. Contributing is also one of the best ways to get hired in the field.