Python 3.10+ · Apache 2.0

AI-First Open Platform for Embodied Robotics

Run the same Python code from simulation to real hardware. 5 lines of code to start your first robot. Under $500 reproducible hardware.

Get Started View on GitHub

navigate_office.py

from threewe import Robot

async with Robot(backend="gazebo") as robot:
    image = robot.get_camera_image()
    await robot.move_to(x=2.0, y=1.0)
    pose = robot.get_pose()

Sim2Real Transfer

Same Python API

<$500 Hardware Cost

5 Lines to Start

4 Sim Backends

12x Less Code vs ROS2

Unified Backend Architecture

A Consistent API, Mock to Real

Switch between physics engines and real hardware with a single parameter — the Python API stays the same across Mock, Gazebo Harmonic, NVIDIA Isaac Sim, and physical robots. Sim-to-real transfer itself still requires task-specific tuning.

bug_report

Mock

Zero dependencies. Pure NumPy 2D kinematic simulation for unit tests and CI/CD.

view_in_ar

Gazebo

Gazebo Harmonic with Nav2 and SLAM. CPU-based physics simulation.

auto_awesome_motion

Isaac Sim

NVIDIA GPU-accelerated parallel RL training with domain randomization.

precision_manufacturing

Real Hardware

Raspberry Pi 5 + ESP32-S3 + Hailo-8L. ROS2 Jazzy at 50Hz control loop.

Core Technology

One API, Four Backends

The same Python Robot class runs against Mock, Gazebo, Isaac Sim, and real hardware — switch with a single parameter. Sensor models and sim configs may still need task-specific tuning; we're upfront about what transfers and what doesn't.

Robot(backend="mock") → Robot(backend="real")

psychology

AI-First API

Designed for transformer architectures. Tokenize sensor inputs and map action spaces directly to VLM/VLA outputs. No ROS2 knowledge required.

account_balance

Open Hardware <$500

Complete BOM, KiCad PCB schematics, STEP models, and assembly guides. Licensed under CERN-OHL-P v2 for full reproducibility.

fitness_center

Gymnasium Ready

Standard OpenAI Gymnasium environments: Navigation, ObjectNav, Exploration, and VLN. Use your existing RL pipelines and Stable Baselines3.

schema

VLM/VLA Integration

Native support for GPT-4o, Qwen-VL, and custom VLA policies. One-line instruction execution: robot.execute_instruction("find the door")

Quickstart Guide

From Zero to Robot in 30s

Install the SDK via pip

Choose your backend (mock for instant start)

Run your first navigation script

Terminal content_copy

$ pip install threewe

... collecting packages

Successfully installed threewe-1.0.4

$ python navigate_office.py

[INFO] Backend: mock initialized

[INFO] Robot active at (0.0, 0.0)

[OK] Navigation complete → (2.0, 1.0)

How 3we Compares

The only platform combining full open hardware with production-grade AI integration.

Feature	3we	TurtleBot 4	ROSbot XL	JetBot
Python API (no ROS)	✓	✗	✗	✗
Same API: mock → sim → real	✓	✗	✗	✗
Open PCB + BOM	✓ Full	Closed	Closed	Partial
Gymnasium Interface	✓	✗	✗	✗
VLM/VLA Native	✓	✗	✗	✗
Hardware Cost	<$500	~$1,200	~$2,800	~$250
Safety (E-Stop)	ISO 13850	Software	None	None

Built for the Ecosystem

science

AI Researchers

Gymnasium environments + VLM integration. Train in Isaac Sim, deploy to hardware in minutes.

school

Students

Learn embodied AI without ROS2 complexity. Full stack from ESP32 firmware to Python policies.

developer_board

RL Practitioners

Standard Gymnasium wrappers, PPO/SAC ready. Benchmark on the public leaderboard.

handyman

Hardware Makers

Fork the open PCB design. PBC-34 hot-plug bus for custom sensor payloads.