Ai2: Building physical AI with virtual simulation data

Virtual simulation data is driving the development of physical AI across corporate environments, led by initiatives like Ai2’s MolmoBot.

Instructing hardware to interact with the real world has historically relied on highly expensive and manually-collected demonstrations. Technology providers building generalist manipulation agents typically frame extensive real-world training as the basis for these systems.

For some context, projects like DROID include 76,000 teleoperated trajectories gathered across 13 institutions, representing roughly 350 hours of human effort. Google DeepMind’s RT-1 required 130,000 episodes collected over 17 months by human operators. This reliance on proprietary, manual data collection inflates research budgets and concentrates capabilities within a small group of well-resourced industrial laboratories.

“Our mission is to build AI that advances science and expands what humanity can discover,” said Ali Farhadi, CEO of Ai2. “Robotics can become a foundational scientific instrument, helping researchers move faster and explore new questions. To get there, we need systems that generalise in the real world and tools the global research community can build on together. Demonstrating transfer from simulation to reality is a meaning...