This AI entrepreneur is developing agents that can plan ahead for the unexpected

Danijar Hafner's new San Francisco startup is still in stealth mode—no name on the door, little furniture—but its open space is filled with humanoid robots imported from China, hanging like marionettes from racks. Hafner describes the venture as a continuation of his long-term research into enabling AI to navigate environments it has not encountered in training. The humanoids are the next, physical step of that work. He argues that overcoming unseen scenarios is key to getting robots into human spaces: a robot sent into a person's home must already be able to handle a floor plan and furniture it has never seen before.

To achieve this, Hafner relies on model-based reinforcement learning. He develops 'world models'—AI models that emulate physical reality—and trains agents within them. The agent treats the model as a real-world simulation, learns how to act there, and uses those experiences to make predictions (or, as he might say, to dream or imagine) about future outcomes. This lets the agents, or the robots they are embedded in, react in unfamiliar situations in real life. Unlike other efforts, his technique enables agents to execute massively complicated tasks without the real-world trial-and-error training traditionally used in robotics.

Hafner grew up in a rural town in northeastern Germany, where his parents were classical musicians. He learned programming from a neighbor and began taking online AI courses in high school. In 2015, as a second-year undergraduate studying engineering at Hasso Plattner Institute, he won a student researcher role at Google Brain, followed by more than a dozen internships and positions across Google Brain and Google DeepMind in the UK, Canada, and the US. He has worked with Geoffrey Hinton and with Ashish Vaswani, coauthor of the groundbreaking 'Attention Is All You Need' paper. His former manager and coauthor Timothy Lillicrap says Hafner easily sits in the top half of 1% of the smart people he interacts with at Google.

The goal is agents that can plan ahead for the unexpected, enabling robots to operate safely in unstructured human environments such as homes. While Hafner remains tight-lipped about his new venture, this approach could be a significant step toward practical real-world robot deployment and other AI applications that must cope with unknown conditions.