The 'OpenAI of Robotics' Endorsed by OpenAI Has Arrived: Valuation Surpasses $2.6 Billion in Less Than Two Years
-
OpenAI and others are 'replicating' an OpenAI in the robotics field, which has become incredibly hot again this year due to AI.
On February 23, 2024, a total of 18 investment firms, including OpenAI, Microsoft, Bezos Expeditions, and NVIDIA, invested $675 million in a robotics company named Figure AI.
Figure AI was founded in 2022. In less than two years, after three rounds of investment, it has almost gathered all the big names in Silicon Valley's tech circle. Moreover, the more you study this company, the more you realize that it follows the same formula as OpenAI did back in the day. Figure AI is focusing on the development of general-purpose humanoid robots. The company's core technology comes from their CTO, Jerry Pratt, who has been researching robotics since 1998. To this day, his research direction has remained unchanged, consistently focusing on the balance and contact (coupling) of bipedal robots.
For example, in 2016, Jerry published a paper titled Walking on Partial Footholds Including Line Contacts with the Humanoid Robot Atlas. The core idea of this paper was to propose a method enabling humanoid robots to walk when footholds are limited.
The advantage of this algorithm lies in not requiring prior knowledge of foothold information while also utilizing expected foothold data to improve stepping effectiveness. After a robot takes a step, it attempts to shift the center of pressure around the edges of its foot to explore new contact surfaces. By rotating the foot around contact edges during exploration and analyzing the actual center of pressure achieved, the robot can infer the available foothold area. This estimated contact region is then utilized by a whole-body momentum-based control algorithm. To walk on incomplete footholds while maintaining balance, the algorithm combines rapid dynamic stepping with upper-body angular momentum for balance recovery. Judging by the title of this paper, you should understand that this algorithm is the core technology behind Boston Dynamics' famous Atlas robot. In fact, without Jerry's algorithm, Atlas wouldn't be able to run and flip so nimbly. If we were to follow Japanese naming conventions, calling Jerry the "Bipedal Robot Sage" wouldn't be an exaggeration at all.
Until 2021, before the concept of collaborative robots (co-robots) became popular, bipedal robots were considered a very "niche" field in robotics. Unlike humans who naturally seek balance on two feet (only considering using hands and knees when physically impaired), robots face much greater challenges with bipedal balance. That's why many robots use three, four, or even eight legs, wheels, or tracks instead. For cases where balance is impossible, they might even be bolted to the ground. This is how the multi-axis robotic arms we know today came into being.
However, as collaborative robots gain popularity and serve as precursors to humanoid robots, researchers working on bipedal robots have become a gold mine of knowledge. Among them, Jerry, who was once active in IEEE, has undoubtedly become the hottest commodity in the eyes of investors. In contrast, Figure AI's CEO Brett Adcock lacks significant academic achievements but is a highly capable businessman with a strong track record in monetization. In 2018, he co-founded Archer Aviation, an electric aircraft company, with Adam Goldstein. In August 2022, United Airlines placed a $10 million deposit to purchase 100 electric air taxis from Archer Aviation. Brett leveraged this deal to take the company public on the New York Stock Exchange.
For those following OpenAI, does the combination of Jerry (likely referring to a technical figure) and Brett seem familiar? Strip away the names, and you have a once-niche technology perfectly suited for the current environment, paired with a businessman who excels at monetizing technology and has an impressive track record—crucially, with Microsoft as an investor.
Doesn’t this sound exactly like OpenAI? Brett corresponds to Sam Altman, though Jerry aligns not with Ilya Sutskever but with Geoffrey Hinton. Figure 01 inserting coffee capsules into the capsule coffee machine
After discussing personnel arrangements, let's talk about the product.
On January 5, 2024, Figure AI released a video showing how they used generative artificial intelligence to train their robot Figure 01 to add coffee capsules to a capsule coffee machine. In the video, a developer standing nearby makes a request to Figure 01, saying: "Can you make me a cup of coffee?" Figure 01 opens the coffee machine lid and inserts the coffee capsule. While the simple action of picking up and inserting the capsule might seem straightforward, the key detail is that when Figure 01 notices the capsule isn't properly positioned, it will reach in to adjust its placement. Figure 01 manually corrects errors when they are detected.
In this demonstration, the unique coupling capabilities of collaborative robots are vividly displayed.
Unlike traditional industrial robots, collaborative robots do not need to be installed within enclosed safety barriers. Instead, they can work closely with the environment without physical isolation. In other words, collaborative robots are not developed for a single specific task, but once developed, they can easily handle a particular task. The technical advantage of Figure 01 lies in its use of generative artificial intelligence to understand instructions by converting natural language into tokens, which are then processed to complete assigned commands. According to the company, they have partnered with OpenAI to jointly develop generative AI technology for humanoid robots.
However, Figure AI's biggest headache also stems from generative AI. As mentioned earlier, Figure 01 understands commands through their corresponding tokens. Yet based on current large language models, there is no method to ensure that generative AI's token output remains consistent and stable during prolonged operation and multiple task switches. In other words, if Figure 01 is made to repeat the same command endlessly, even if it has performed this action countless times, it will inevitably make an error at some point - this is essentially the "Shakespeare monkey" theory.
Now that we understand the company's core team and operations, when we re-examine Figure AI's investments, you'll notice that "something smells off." Figure AI was founded in 2022, and this financing round marks the company's third. However, what's remarkable is that this company, having gone through only three rounds of financing, has managed to attract a staggering 28 investors. Moreover, the company's financing wasn't a complete three rounds; one of them was an additional investment from Intel and Big Sky Partners. In reality, Figure AI has only experienced two rounds of investment.
For those familiar with investments, it's generally understood that in corporate financing, unless there are many relatives and friends involved, it's rare to see a large gathering of individual investors in the early rounds. Early-stage financing typically involves a limited number of investors, mainly due to factors such as high risk, valuation uncertainty, information asymmetry leading to cautious investment attitudes, and the constraints of the startup's own needs and practical conditions. Since the amounts raised in early rounds are relatively small, the pool of qualified investors fitting this scale is inherently limited. Additionally, any financing requires handling extensive due diligence, legal documentation, and equity registration for numerous investors, which is time-consuming and labor-intensive. For early-stage entrepreneurs, expending too much energy on these matters can severely hinder company development. Brett, in this regard, should be well-versed and unlikely to make mistakes. Moreover, it's not just about the sheer number of investors. As mentioned at the beginning of the article, Figure AI's shareholders are not only numerous but also incredibly influential. When these big names speak, the world listens.
This indicates one thing: these massive capital injections are not 'investments' but rather 'orders'.
The robotics industry is exceptionally unique—it's an extremely complex interdisciplinary field. Upstream, it involves metals, rare earth elements, and energy; midstream includes precision components, optical parts, software, and chips; downstream features the most representative applications of humanoid robots in logistics, services, and exploration. These investors are essentially placing orders with Figure AI through their investments, thereby reaching the entire upstream, midstream, and downstream sectors of robotics. This is an industry that demands stronger resource integration capabilities than the AI field, and today Brett from Figure appears to have been anointed by the big shots as the next Altman-like figure. A new OpenAI-style story is on the horizon.