Sudo AI introduces simulation-trained Sudo R1
Sudo AI introduced Sudo R1, an integrated robot system for object picking trained entirely in simulation.
Evidence notes
- Sudo R1 combines company-developed hardware and software with a manipulation-focused foundation model.
- The system uses no real-world demonstrations or manual labels for training before deployment to physical hardware.
- A 60-minute uncut evaluation reported about 98% first-attempt picking success across unfamiliar objects and changing conditions.
Company context
Sudo AI develops foundation models for robotic manipulation, trained in large-scale simulated environments before deployment on physical hardware. Its first public system, sudo R1, combines self-developed hardware and software around object picking as a core manipulation primitive.