Source of article reprint: Silicon Star People
Article source: Silicon Star Pro
Author of the article: Miao Zheng
图片来源：由无界AI生成
OpenAI is "replicating" OpenAI in the field of robotics, which has become extremely popular again due to AI this year. On February 23, 2024, a total of 18 investment companies including OpenAI, Microsoft, Bezos Venture Capital, and NVIDIA invested US$675 million in a robotics company. This company is Figure AI. Figure AI was founded in 2022. After three rounds of investment in less than two years, it has gathered almost all the prominent figures in the Silicon Valley technology circle. Moreover, the more you study this company, the more you will find that this company and OpenAI have the same recipe.
Figure自己的Altman+Ilya组合
Figure AI is working on a general-purpose humanoid robot. The company's core technology comes from their CTO, Jerry Pratt. He has been studying robots since 1998, and to this day, the direction of his research has never changed. It has always been the balance and contact (coupling) of bipedal robots. For example, in 2016, Jerry published a paper "Walking on Partial Footholds Including Line Contacts with the Humanoid Robot Atlas". The core idea of this paper is to propose a method for humanoid robots to walk with limited footing. The advantage of this algorithm is that it does not need to know the foothold information in advance, and it can also use the information of the expected foothold to improve the stepping effect. After the robot takes a step, it attempts to explore new contact surfaces by changing the position of its center of pressure around the edge of the foot. The available footing area is inferred from the way the foot rotates around the contact edge during exploration, and the actual center of pressure position achieved. This estimated contact area is then exploited by a whole-body momentum-based control algorithm. To walk and maintain balance on incomplete footings, the algorithm combines fast and dynamic steps with the use of upper body angular momentum to restore balance. Then you should understand from the title of this paper that this algorithm is the core algorithm used by Boston Robotics’ popular Atlas. In fact, if it weren't for Jerry's algorithm, Atlas wouldn't be able to run and flip like a rabbit. If we go by the Japanese style, it would not be an exaggeration to call Jerry a bipedal robot immortal.
Atlas In fact, until 2021, before the concept of collaborative robots (co-robots) became popular, bipedal robots were a very "alternative" field in the field of robotics. Because robots are different from us humans. When humans pursue balance, they can only consider their feet. Only those with physical impairments will consider their hands and knees. But the robot is different. It is difficult to control the balance of the two legs. Then instead of using two legs, use three legs, four legs, eight legs, wheels, and tracks. If it is not balanced, just put it on the ground with screws. This is how the multi-axis robotic arm we know today came about. But as collaborative robots become more popular, scholars of these bipedal robots, the precursors to humanoid robots, have become a gold mine. Jerry, who was once active in IEEE, is undoubtedly the most popular among capital. In comparison, Figure AI CEO Brett Adcock has no academic achievements. He is a businessman with strong liquidity. In 2018, he and his partner Adam Goldstein founded the electric aircraft company Archer Aviation. In August 2022, United Airlines paid a $10 million deposit to Archer Aviation to purchase 100 electric flying taxis. Brett directly listed the company on the New York Stock Exchange through this order. Friends who follow OpenAI, do you think the combination of Jerry and Brett looks familiar? We took off the name and said that there was such a technology that was very unpopular in the past, but is very suitable for the current environment, plus a businessman with strong technology monetization capabilities and an impressive record. The key is that the capital of this company has Microsoft. Isn’t this OpenAI? Brett corresponds to Ultraman, but Jerry corresponds not to Suzkowe, but to Hinton.
Figure 01 Add capsule coffee to the capsule coffee machine
The robot with the strongest sense of large model
After talking about staffing, let’s talk about products. On January 5, 2024, Figure AI released a video in which they used generative artificial intelligence to train Figure AI’s robot Figure 01 to add coffee capsules to and from the capsule coffee machine. In the video, the developer made a request to Figure 01 from the side. He said to Figure 01: "Can you make me a cup of coffee?" Figure 01 opened the lid of the coffee machine and added the coffee capsules into the coffee machine. It would be fine if you just pick up the coffee capsule and put it into the coffee machine. The key is that if Figure 01 finds that the coffee capsule is not placed in the right position, he will also reach out to straighten the coffee capsule.
Figure 01 After an error is discovered, it will be corrected manually. In this demonstration, the unique coupling ability of the collaborative robot is fully demonstrated. Collaborative robots are different from traditional industrial robots. Collaborative robots do not need to be installed in a closed safety fence, but can work closely with the environment without physical isolation. In other words, collaborative robots are not developed to do a certain thing, but the collaborative robots developed can easily do a certain thing. The technical advantage of Figure 01 is that the way it understands instructions is to use generative artificial intelligence to convert natural language into tokens, and then process the tokens to complete the arranged commands. According to the company, they have partnered with OpenAI to jointly develop generative artificial intelligence technology for humanoid robots. However, the biggest headache for Figure AI currently comes from generative artificial intelligence. As we just said, the way Figure 01 understands instructions is the token corresponding to the instruction. However, according to existing large language models, there is no way to ensure that the output of tokens by generative artificial intelligence is consistent and stable during long-term running and multiple task switching processes. In other words, if Figure 01 is asked to repeat the same command, even if it repeats this action countless times, it will definitely make an error at some point. This is the case with the "Shakespeare's Monkey" theory.
What is put together is not financing, but...
Now that we understand the company's core personnel and business, when we look back at Figure AI's investment, you will find that "it doesn't taste right." Figure AI was founded in 2022, and this financing is the company’s third financing. But a magical thing is that this company, which has only gone through three rounds of financing, has 28 investors. Moreover, this company’s financing is not a complete three rounds. The middle round is additional investment from Intel and Big Sky Partners. So in fact, Figure AI has only experienced two rounds of investment. Friends who often invest know that generally speaking, when it comes to corporate financing, unless there are many relatives and friends, only in this case will there be a large number of individual investors with their families, otherwise it will rarely happen in the first few rounds. A large number of investors. In the first few rounds of financing, the number of investors was generally limited, mainly due to cautious investment attitudes caused by factors such as high risk, uncertainty in valuation, and information asymmetry, as well as the needs and practical conditions of start-up companies themselves. Because the financing amounts in previous rounds were relatively small, the pool of qualified investors suitable for this size of investment is inherently limited. Moreover, any financing needs to handle a large number of investors' due diligence, legal documents, equity registration and other tasks, which is very time-consuming and labor-intensive. For early-stage entrepreneurs, wasting too much energy on these matters will seriously affect the development of the company. Brett should be very familiar with this aspect and cannot make mistakes. And it doesn’t matter if there are many investors. As mentioned at the beginning of the article, Figure AI’s shareholders are not only large in number, but also have terrifyingly large positions. When these big guys roar, the earth will shake three times. This only shows one thing, that is, these huge capital injections are not "investments", but "orders". The robotics industry is very special. It is an extremely complex interdisciplinary subject. The upstream includes metals, rare earths, and energy, the midstream includes precision parts, optical parts, software, and chips, and the downstream includes the most representative logistics, services, and exploration of humanoid robots. These investors place orders for Figure AI through investors. Then reach the upstream, midstream and downstream of the robot through orders.This is an industry that requires stronger resource integration capabilities than the AI field, and today Figure’s Brett seems to have been designated by the big guys as the next Altman-like figure. A new OpenAI-style story is coming.
