Behind the huge popularity of Figure 03: A well-choreographed "concept device" frenzy?

On October 10, 2025, Brett Adcock, a rising star in the field of American humanoid robots, "dominated" social media with a video: His latest product, Figure 03, served tea, did laundry, and took care of children in a simulated home environment. Its natural movements made it seem as if it had perfectly integrated into human life.

However, beneath this viral video lies a covert battle over technological authenticity and the future of the industry.

According to official data, compared with Figure 02 released in August 2024, Figure 03 is 9% lighter and significantly smaller in size. The key parts are designed with multi - density foam to prevent pinching, and the outer layer is covered with soft fabric, replacing the previous hard mechanical shell, making it easier to move in a home environment.

On the day of the release, founder Brett Adcock posted on X: "By the way, nothing in this film is teleoperated."

This statement seemed to be specifically addressed to someone.

Just a year ago in October, Elon Musk held a press conference titled "We, Robot" in California. At the conference, the Optimus robot mixed drinks for guests on stage, greeted people at the scene, and even danced a "group dance" to the music, looking very "human - like".

After the conference, a guest revealed that the Optimus robot on site was actually remotely controlled. However, Musk did not disclose any technical details or respond afterwards.

The statement Brett Adcock posted on X was intended to prove that Figure AI can make humanoid robots stop being "puppets" of humans and enable them to move independently.

However, the truth may have another side.

01 Figure 03 Became an Internet Sensation for Its Perfect Integration into Family Life

The last "household robot" to gain popularity was NEO, a humanoid robot under the Norwegian company 1X Technologies invested by OpenAI. In the official video released in September 2024, NEO tidied up glasses, sorted tableware in the kitchen, and even hugged humans. Its gentle and smooth movements gave people the first intuitive impression that robots can do housework.

After the release of Figure 03, Bernt Bornich, the founder of 1X Technologies, made a veiled sarcastic comment on social media: "Imitation is still the sincerest form of flattery, but taste is hard to copy."

Obviously, he was subtly mocking that the appearance and concept of Figure 03 were too similar to those of NEO.

The popularity of Figure 03 and NEO is not because they completed a certain complex task, but because they demonstrated a beautiful vision of robots perfectly integrating into human family life. This indicates that humanoid robots have reached a singularity moment from "laboratory demonstrations" to "coexistence in the real world".

According to the official website of Figure AI, three key upgrades contribute to this effect: a smarter "brain", a more flexible pair of "hands", and a body that can be "dressed" for the first time.

First, the upgrade of the brain.

In early 2025, Figure AI terminated its cooperation with OpenAI and started to develop its own VLA model, Helix.

Figure AI revealed that currently, teaching a robot a new movement often requires engineers to write scripts by hand or provide thousands of demonstrations, which is costly and inefficient. The goal of Helix is to directly transform the "common sense" (recognizing objects in pictures, understanding relationships and instructions) learned by the vision - language model (VLM) through large - scale training on the Internet into actions. It adopts a "dual - system" design with a vision - language - action (VLA) architecture:

● S2 is an on - board VLM (about 7–9 Hz), responsible for "slow thinking" — understanding the scene and language, and transforming "what to do" into a transmissible semantic representation;

● S1 is a 200 Hz vision - motion strategy, responsible for "fast reaction" — translating the semantic representation of S2 into continuous joint/finger control in real - time and making immediate corrections according to sensory feedback.

With end - to - end joint training and communication between the two layers, the robot no longer runs pre - set scripts, and many new tasks can be learned with just natural language prompts.

In order to obtain sufficient data, Figure AI launched Project Go - Big in September this year. The plan is to build the world's largest pre - training dataset for humanoid robots using the scene data collected by hundreds of Figure 03 robots, providing continuous real - world feedback for Helix.

Second, the progress of the dexterous hands and feet.

Figure 03 wears gloves throughout, so it's difficult to see the details. However, from the demonstration, each palm of Figure 03 has a built - in small camera, which can provide close - range redundant vision when the main camera is blocked. Moreover, each fingertip of Figure 03 can detect a force change as small as 3 grams — figuratively speaking, it can sense the force of a paperclip.

The softer fingertip material increases the contact area, enabling Figure 03 to stably grasp objects of different shapes and hardness. In the video, the "stability and gentleness" shown when picking up a computer and holding a cleaning ball is the result of the combination of tactile and control systems.

The foot structure of Figure 03 has been significantly improved compared with the previous generation. It uses a hollowed - out arch and independently movable toe structure, which can fine - tune the support point and landing angle while walking, allowing for a smoother center - of - gravity transfer between steps. This design concept is similar to the mechanical foot structure of Tesla's Optimus, aiming to make the gait more stable and closer to that of humans.

In addition, the logo "Inductive Charging" on the instep indicates that this robot supports inductive wireless charging. It can automatically recharge when standing still, eliminating the need for manual plugging or lying down for charging, leaving a convenient interface for future continuous - work scenarios.

Finally, the integration of body changes and home scenarios. Figure 03 wears a light - gray fabric coat with a soft surface, and the joint lines are hidden as much as possible. The official has not released specific material information, but from the actual performance, it can work for a long time while wearing the covering layer, indicating that Figure AI has full confidence in its heat dissipation and energy - efficiency control.

In the past, due to high motor thermal power and difficult heat dissipation, humanoid robots usually had exposed metal structures. However, Figure 03 can operate stably in an enclosed space, indicating that it has reached the level of heat management and energy efficiency to support home use, which is very similar to the NEO robot a year ago.

This not only changes the appearance of humanoid robots but also shortens the psychological distance between humans and them — humanoid robots can make people no longer afraid to get close. The popularity of Figure 03 is not because it is smarter, but because it is more like a trustworthy presence. It doesn't show superpowers, but makes people believe that machines can move around at home without causing unease.

Globally, the technological routes of humanoid robots are gradually diverging: some are pursuing "the limits of the body", while others are exploring "the budding of intelligence".

Boston Dynamics' Atlas is a representative of the former. The early hydraulic version achieved running, climbing, and jumping with high - power drive, and the latest electric version can even perform 360 - degree flips and coordinated full - body weightlifting. It represents the peak of motion control but is mainly used in industrial, emergency, and scientific research scenarios, still far from daily life.

Digit, invested by Amazon, goes to the other end — it is more like a warehouse worker, focusing on package handling, going up and down stairs, and collaborating with automated systems. The focus of Digit is not "showing off skills" but stability, reliability, and safety.

In China, the iteration speed of humanoid robots is also very fast. Unitree Robotics continuously optimizes the body structure and gait algorithm through products such as H1, G1, and R1, emphasizing lightweight and mass - production potential; Zheyuan's "Lingxi" and "Yuanzheng" series focus on entertainment and educational interaction; Zhuji Dynamics' Oli can complete a complete behavior chain such as tennis recognition, ball - picking, and placement; the rapidly evolving Booster T1 can switch between autonomous and remote - control modes in football scenarios. They jointly promote China's engineering progress in the fields of electromechanical systems and motion control.

In the product demonstrations of humanoid robots at home and abroad, Figure 03 is making faster progress towards home scenarios through details such as tidying up pillows, moving express deliveries, and avoiding pets.

02 Although It's an Internet Sensation, It's Still a Conceptual Product

Although Figure 03 has become popular, it is still only at the demonstration level and far from entering real human life. This conclusion is mainly drawn from two aspects of observation:

One is from the on - site experience of a reporter from Time magazine, and the other is from the technological reality of Figure AI itself.

In August 2025, a reporter from Time magazine visited Figure AI's headquarters in Silicon Valley. At that time, Figure 03 was still in the closed - training stage, and what the reporter actually saw was Figure 02, the previous - generation robot product with the same software.

The reporter said that in the demonstration of "the robot putting clothes into the washing machine", the machine dropped the clothes on the ground twice in a row and only succeeded on the third attempt.

It can be seen that its "natural and smooth" performance does not come from autonomous intelligence but from repeated training and parameter adjustment. The seemingly one - take video is probably the smoothest attempt out of dozens of shootings.

Brett Adcock said in an interview that the robot still has obvious limitations. He emphasized that "the problems are not difficult to solve" and explained that the neural network Helix of Figure robots can learn new tasks with a very small amount of data — for example, the ability to fold towels was learned through only 80 hours of video training. He also believes that "data can solve all current problems", and as more data is input, the robot's operation speed will increase and errors will decrease.

However, obviously, what he calls "occasional errors of the robot" and "the error rate has dropped significantly and is almost zero" do not match the facts seen by the reporter.

In addition, on Hacker News, a technology community operated by the startup incubator Y Combinator, some practitioners in the humanoid robot field made very straightforward comments:

● "All the videos are carefully selected."

● "If you let it repeat the action ten times, most attempts will fail."

● "These demonstrations are not scientific achievements but marketing works."

● A developer who used to work in a robot company said bluntly: "Ninety percent of today's robot demonstration videos are perfect shots, and there is even a lot of falsification. The only real standard is whether you can buy it and use it. If not, it's probably just a financing public - relations stunt."

An Amazon engineer who has long been concerned about the humanoid robot field told the author: "This release is probably prepared by Brett Adcock for financing."

The fact is that in the past year, the financing rhythm of Figure AI has been almost synchronous with its product release rhythm.

In February 2024, Figure AI announced that it had received joint investment from many technology giants such as Amazon founder Jeff Bezos, Microsoft, Nvidia, and OpenAI, completing a Series B financing of approximately $675 million. After this round of financing, the valuation of Figure AI soared to $2.6 billion.

In September 2025, Figure AI completed a Series C financing of $1 billion, and the post - financing valuation of the company rose to $39 billion, making it the humanoid robot company with the highest valuation globally.

Founder Brett Adcock is an experienced entrepreneur. He is very good at "finding people" and "finding money". Before founding Figure AI, he created the recruitment platform Vettery and the electric aircraft company Archer Aviation, attracting investment from Walmart executive Marc Lore.

In 2022, when GPT - 3 was on the rise, he left Archer and turned to the humanoid robot field, founding Figure AI and inviting robot scholar Jerry Pratt and DeepMind scientist Corey Lynch to join.

In April 2023, after Figure