Open Source for Awkward Robots

In this episode Jan Lipart, CEO of OpenMind, explains how his team is creating an open‑source operating system—OM1—for humanoid robots. Drawing on a background in physics, healthcare AI, and hobbyist robotics, Lipart highlights the convergence of large language models (LLMs) with physical agents. He argues that LLMs can generate actionable commands for hardware, and that embedding immutable Asimov‑style safety rules on Ethereum provides a transparent governance layer for autonomous machines.

The technical architecture revolves around natural‑language communication between subsystems. Vision, battery, and inertial models each output plain sentences, which are fused into paragraphs and fed to a decision‑making LLM. A supervisory “coach” model periodically corrects the robot’s behavior, acting like an internal referee. To simplify hardware integration, OpenMind attaches a standardized brain‑pack—typically an NVIDIA Orin paired with RealSense sensors—to any commercial humanoid, dramatically shrinking the driver matrix. This modular approach lets developers focus on high‑level cognition rather than low‑level driver code.

Beyond the stack, OpenMind envisions a decentralized app store where developers publish skill modules, turning robots into programmable extensions of everyday devices. By keeping the software stack open, the company aims to avoid opaque, over‑the‑air updates that dominate consumer robotics today. Lipart stresses that societal acceptance, regulation, and liability will become the next frontier as robots enter homes, schools, and workplaces. Open source transparency, combined with blockchain‑anchored safety policies, could bridge the trust gap and accelerate the adoption of human‑focused robots in the coming decade.