Why Embodied AI Requires Robotics to Be Open All the Way Down

For two decades, open robotics meant ROS—shared middleware that let researchers avoid reinventing message-passing. But today, Embodied AI requires the entire stack to be open: from PCB schematics to Python APIs. Most platforms are open only at the middleware layer, leaving hardware and firmware as black boxes. The 3we platform, a fully open robot platform, demonstrates what 'open all the way down' means in practice: Apache-licensed Python SDK, CERN-OHL-P hardware documentation, and a ROS2 stack where the same code runs identically in simulation and on real hardware.

Three forces make full-stack openness non-negotiable: the sim2real bottleneck (which requires control over firmware and sensor models), the need for clean async-first APIs to connect foundation models to realtime control loops, and a reproducibility crisis where most papers cannot be independently verified. 3we's design is 'AI-first'—its primary API is a Python async class, not raw ROS2 topics—allowing researchers to write code like `await robot.move_to(x=2.0, y=1.0)` without ever touching the underlying hardware details. This lowers the barrier for multi-agent experiments (ten robots at $500 each) and enables true reproducibility through published BOMs.