Moonlake AI

Moonlake AI is a frontier AI lab building the evaluation infrastructure for robot deployment. Robots learn slowly because real-world data is scarce and expensive to collect. Moonlake’s answer is simulation that is accurate enough to stand in for the real thing.

Its platform generates physics-ready digital assets and environments from unstructured input: a photo, a video, or a text description. The outputs come with articulated joints, real-world ranges of motion, and drive limits, and export directly to NVIDIA Isaac Sim, MuJoCo, and USD without manual cleanup. On top of that, Moonlake offers sim-real matched evaluation, so a team can measure how closely a simulated test predicts hardware performance, and policy improvement tools that use targeted simulation data and reinforcement learning to raise robot throughput on specific tasks.

Early use cases span electronics handling and inspection, packaging and logistics, and service tasks in commercial spaces. The company is based in San Francisco. Co-founders Fan-Yun Sun and Sharon Lee are Stanford AI PhDs; Sun previously led 3D generation research at NVIDIA. The team draws from DeepMind, NVIDIA, Tesla, Epic Games, Waymo, and Stability AI, and includes Chris Manning, formerly of Stanford, as Distinguished MTS.