Google’s 7-Year Quest to Equip AI with a Robot Body

In early January 2016, I had just joined Google X, Alphabet’s secretive innovation lab, tasked with determining what to do with the technology and teams from nine robot companies Google had acquired. This project followed the sudden and mysterious departure of Andy Rubin, the renowned "father of Android," leaving many at the lab unsure about the future. Larry Page and Sergey Brin occasionally dropped in to offer guidance, but it was Astro Teller, the head of Google X—affectionately referred to as the "moonshot factory"—who had decided to bring the various robot teams into the lab just a few months earlier.

I was drawn to Google X because Astro convinced me it would be different from other corporate innovation labs. The lab was grounded in the founders' commitment to audacious thinking and backed by Google’s "patient capital." After a career spent founding and selling several tech companies, joining X felt like a natural fit. X seemed to embody the sort of big bets that a company like Google should make. The idea of building robots that would one day work and live alongside humans was one such bet.

Fast forward eight and a half years, and although Google has discontinued its largest robotics project, new robotics startups seem to spring up every week. I am more convinced than ever that robots are essential to our future. Yet, I worry that Silicon Valley’s focus on “minimum viable products” and the venture capital world’s reluctance to invest in hardware will hinder progress. Moreover, much of the current investment seems to be targeting the wrong priorities.

Google X, where our project—eventually named Everyday Robots—was housed, was launched in 2010 with a mission to tackle some of the world’s hardest problems. The lab was located in its own building, several miles from Google’s main campus, to foster a unique culture and encourage moonshot thinking. This meant taking big risks, rapidly experimenting, and celebrating failure as a sign of having set ambitious goals. By the time I arrived, X had already given birth to some futuristic projects like Waymo, Google Glass, and other seemingly science-fictional endeavors such as flying wind turbines and balloons designed to provide internet access.

What set X apart from traditional Silicon Valley startups was the sheer scale and long-term nature of the projects. To be classified as a "moonshot," an idea had to meet three criteria: it had to solve a problem affecting hundreds of millions of people, rely on breakthrough technology, and propose a radical solution that was just on the right side of crazy.

Astro Teller, fittingly titled the "Captain of Moonshots," perfectly encapsulated the culture at X. You would never find him without his rollerblades, ponytail, and an ever-present friendly smile. He was exactly the kind of leader a place like Google X needed.

When Astro and I first discussed the future of the robotics teams, we knew we had to do something big—but what exactly? At the time, most robots were large, dangerous machines confined to factories and warehouses, far from being suitable for everyday environments. To tackle this, we focused on the growing global challenges of aging populations, shrinking workforces, and labor shortages. Our breakthrough technology would be artificial intelligence, and our radical solution was to create fully autonomous robots capable of assisting with daily tasks.

In essence, we aimed to give AI a body—a robot capable of existing in the physical world. This vision would require patience, experimentation, technical breakthroughs, and likely billions of dollars. Despite the challenges, the team was united in the belief that the convergence of AI and robotics was inevitable, and that much of what had previously been relegated to science fiction was on the cusp of becoming reality.

Each week, my mother would call and ask, “When are the robots coming?” She was living in Oslo, Norway, and while she had good public healthcare and caregivers helping with her advanced Parkinson’s disease, she longed for robots to assist with small, everyday tasks. These were the kinds of tasks that, while simple to most, had become insurmountable barriers for her. It was a deeply personal reminder of the need for robots to make everyday life easier for people like my mother.

As we progressed, we quickly learned that robotics is inherently a systems problem—a robot is only as good as its weakest component. Jeff Bingham, a bioengineering PhD and a core member of our team, reminded me of this often. Every robot subsystem—whether vision, navigation, or grasping—had to function seamlessly in unpredictable environments. For decades, even simple tasks like picking up a cup had proven incredibly difficult for robots because the real world is so unpredictable.

It became clear that we needed to approach AI in robotics in one of two ways: a hybrid model, where some tasks were powered by AI while others used traditional programming, or an end-to-end learning (e2e) model, where robots learned entire tasks, much like humans do. While the hybrid model had its merits, the real breakthrough lay in e2e learning. Larry Page, with his characteristic cryptic insight, once told me that we only needed 17 machine-learning experts to solve the problem. At first, I didn’t understand his point, but I eventually realized that he was emphasizing the importance of small, focused teams making big breakthroughs—not the actual number.

Despite Larry’s confidence, building robots that could reliably perform tasks in the real world was no small feat. But we were determined to crack the code of AI-powered robotics. In the process, we sought to build robots that could operate in the messy, unpredictable environments of everyday life—a challenge that would ultimately define our moonshot at Google X.