Former Google researcher uses model-based reinforcement learning to prepare robots for spaces they have never seen.
AI researcher Danijar Hafner is building a stealth startup focused on agents that can plan for unexpected environments. His San Francisco office sits mostly empty, but humanoid robots hang from racks throughout the space. Hafner imports the robots from China and trains their agents with model-based reinforcement learning inside simulated world models.
Hafner, 31, grew up in northeastern Germany and studied engineering at Hasso Plattner Institute. He joined Google Brain as a student researcher in 2015 and later held roles at Google Brain and Google DeepMind across the UK, Canada, and the US. His collaborators there included Geoffrey Hinton and Ashish Vaswani, coauthor of the transformer architecture.
The training method lets robots rehearse many physical scenarios before they encounter an unfamiliar space. The approach removes the need for expensive real-world trial-and-error in robot training. Operators gain a practical path to deploy humanoids in homes and offices, provided the simulated world models capture enough real-world variability.
Hafner’s startup remains in stealth mode with no public name on its San Francisco door. The humanoid robots hanging in his office are a physical embodiment of his world-model work. The next public milestone will be evidence that these agents can operate in novel homes and offices. The result will show whether simulated training can take robots from office racks into everyday spaces.
What matters
- Hafner’s startup trains agents in simulated world models instead of using real-world trial and error.
- The method reduces the need for real-world robot training in unfamiliar homes and workplaces.
- Watch for Hafner’s stealth startup to reveal humanoid robots capable of handling novel floor plans.
Why it matters
Watch for Hafner’s stealth startup to reveal humanoid robots capable of handling novel floor plans.
This GenAI News article was prepared in original wording using reporting and materials published by MIT Technology Review AI. Source reference: https://www.technologyreview.com/2026/09/08/1142088/danijar-hafner-developing-plan-ahead-agents/.
Drafted by the GenAI News review pipeline.