Sudo AI raises more than $149M to scale simulation-trained robots

Sudo AI builds software and hardware that let robotics teams train manipulation skills in simulated environments before putting them on physical machines. Dealroom reported that the Shanghai company raised more than $149 million in a late-stage round announced September 30.

Sudo says it creates robot practice data in a virtual environment, trains a control policy on those simulated objects and conditions, and then transfers the policy to physical hardware. That avoids recording human demonstrations on real machines for every object or scene, while giving the model more lighting, background and obstacle combinations to practice against.

The company’s R1 system packages the approach into its own robot and model, starting with object picking. Sudo reports about 98% first-attempt success and nearly 100% within two attempts across unfamiliar objects and changing conditions. Those are company evaluations, and Sudo says production-grade performance still lies ahead.

The initial target isn’t limited to factories buying finished machines. In a July interview with the Association for Advancing Automation, co-founder and operating chief Martina Hansen said researchers were Sudo’s biggest target market and described a developer platform spanning simulation, data and infrastructure for third-party robotics systems. She also said the company expected to begin taking robot preorders.

Shanghai Guotou Pioneer Fund, Guotai Haitong and Futeng Capital co-led the financing, while existing shareholder Xiaomi invested again, according to Dealroom. Sudo said the capital will support further full-stack development and larger-scale testing in real industrial settings.