Figure 01 walked and the humanoid robot race

Figure AI released a video of their Figure 01 robot walking. Upright. Balanced. Smooth enough that you wouldn’t describe it as stumbling.

It’s not the most impressive walking robot. Boston Dynamics’ Atlas is still the gold standard for dynamic locomotion. Tesla’s Optimus is improving steadily. Agility Robotics’ Digit has been deployed in warehouses.

But Figure’s announcement matters because of timing. Not what the robot can do, but when it appeared, and alongside whom.

The lineup

As of today, at least five well-funded companies are actively developing humanoid robots intended for commercial deployment:

Boston Dynamics: Atlas, the most capable bipedal robot in existence. Research platform. Not sold commercially as a humanoid (Spot, the quadruped, is the product line). But Atlas development continues.

Tesla: Optimus. Walked on stage in October 2022. Improving rapidly based on leaked videos. Tesla’s manufacturing scale is the differentiator here.

Agility Robotics: Digit. Already deployed in Amazon warehouses for testing. Focused on logistics tasks. The most “practical” of the group.

Unitree: H1. A Chinese company that’s been making quadruped robots (the Go2 is a viral video staple) and recently showed a humanoid prototype.

Figure AI: Figure 01. Fresh. Well-funded ($70 million Series A). Led by Brett Adcock, who previously founded two other companies.

Five companies. Different approaches. Different timelines. Different target markets. But all converging on the same form factor: two legs, two arms, a torso, a head. Human-shaped.

Why the timing matters

Three years ago, the idea of a humanoid robot startup was a punchline. The hardware was too hard. The software was too primitive. The business case was too uncertain. Boston Dynamics had been working on it for decades with DARPA funding and still hadn’t found a commercial market for humanoids.

What changed? Two things.

First, AI improved. Large language models and computer vision models made it possible for robots to understand instructions and interpret environments in ways that weren’t feasible before. A humanoid robot is only useful if it can understand what you want and perceive the space it’s operating in. LLMs and vision models are closing that gap.

Second, the economic incentive crystallized. Labor shortages. Rising wages. An aging population in most developed countries. The economic case for a robot that can do physical work in human environments just got stronger.

What I’m watching

The race isn’t about who builds the best robot. It’s about who builds the first robot that’s good enough. Good enough to carry a box in a warehouse. Good enough to stock a shelf. Good enough to load a dishwasher. The bar for “useful” is much lower than the bar for “impressive.”

Digit is closest to useful right now. But Tesla has manufacturing scale. Figure has fresh funding and the advantage of starting from scratch. Unitree has speed and cost competitiveness.

I’ve been following bipedal robots since the DARPA Robotics Challenge in 2015, when the best robots in the world couldn’t reliably open a door. Eight years later, five companies are racing to commercialize the form factor.

The race is on. I don’t know who wins. But the fact that there’s a race, with multiple well-funded entries, tells me the market believes the destination is real. And I’ve learned to pay attention when the market believes something.

Related thinking: