AI Startup Founder Builds Robots That Plan for the Unexpected
Danijar Hafner, a former Google DeepMind researcher, has launched a stealth startup in San Francisco using humanoid robots and world models to help AI navigate new environments.
Hafner, who left Google DeepMind in fall 2025, builds "world models" that emulate physical reality and trains AI agents inside them. These agents learn to act in simulated environments and then use those experiences to predict future outcomes, allowing robots to handle floor plans, furniture, or obstacles they have never encountered before. Unlike traditional robotics methods that rely on real-world trial and error, Hafner's approach lets agents execute complex tasks without direct physical training.
Hafner grew up in rural northeastern Germany, where his parents were classical musicians. He learned programming from a neighbor and later studied engineering at Hasso Plattner Institute in Potsdam. In 2015, as a second-year undergraduate, he became a student researcher at Google Brain, followed by a dozen internships and positions at Google Brain and Google DeepMind in the UK, Canada, and the US. He worked with Geoffrey Hinton, often called a godfather of AI, and Ashish Vaswani, coauthor of "Attention Is All You Need," the paper that introduced transformer technology.
His former manager at Google, Timothy Lillicrap, described Hafner as among the top half of the top 1% of researchers, noting that he often built single-handedly what would take entire teams of engineers. Over the years, Hafner demonstrated his approach with a series of projects. His PlaNet model allowed agents to plan actions ahead. Dreamer 2 became the first world-model agent to reach human-level performance in Atari 2600 games. Dreamer 3 was the first to solve the Minecraft Diamond challenge, and Dreamer 4 learned to mine diamonds from recorded gameplay videos without ever interacting with the game directly. His DayDreamer project brought the Dreamer algorithm into physical robots that could react to new experiences, such as being pushed over, without specific training.
Hafner remains coy about the details of his new venture, but he describes it as a continuation of his work on enabling AI to navigate environments unseen during training. "I was interested in solving a problem," he hinted, "that would change the world."