Speech-to-speech inference functionality. This feature has been fully upgraded in Figure 02.
Figure 02 is equipped with microphones and speakers, leveraging OpenAI's power to achieve voice dialogue and reasoning.
In the technical article published by founder Brett Adcock, he introduced how Figure 02 turns ChatGPT into a robot:
Users input speech, Figure 02 converts speech to text, while ### the newly added 6 RGB cameras give the robot superhuman vision, able to receive image information. Both types of information are processed by ChatGPT.
The AI-processed information is fed back to users in the form of speech and guides the robot to take responsive actions.
Feedback alone is not enough; during actual execution, it needs to work with visual models. Otherwise, there would be mishaps like the robot spilling all the food in the pan onto the ground while cooking. Figure 02 has put a lot of effort into this.
Adcock introduced that ### Figure 02 has a built-in Vision Language Model (VLM) working with 6 cameras, allowing the robot to perform semantic-based and rapid common-sense visual reasoning.
This functionality is fully demonstrated in the collaboration with BMW.
In January this year, Figure AI announced a commercial agreement with BMW Manufacturing Co., LLC to deploy general-purpose robots in automotive manufacturing processes. The promotional video for Figure 02 also dedicates a significant portion to showcasing how Figure 02 uses the vision language model for precise component assembly work in BMW factories.
Moreover, compared to the previous generation, the AI reasoning capability has improved threefold. In the promotional video, Figure 02 is shown fixing components that were not properly installed.
Besides these, what people care about most is the improvement of the robot's "hands".
Our hands can easily count from 1 to 10. But such a simple gesture is extremely difficult for robots. When the teaser was released, everyone's attention was drawn to the fourth-generation hands.
This time, Figure 02's ### fourth-generation hands are equipped with 16 degrees of freedom.
The number "16" means 5 fingers, with 3 joints on each finger, totaling 15 joints, plus 1 wrist joint, making 16 joints that can move freely. This technology is a big step closer to the human hand's 22 degrees of freedom.
Additionally, Figure 02's hands are equipped with strength comparable to humans, capable of bearing 25 kg, making them more suitable for practical scenarios. Some netizens have posted comparisons between 01 and 02, showing that with the increase in degrees of freedom, hand movements appear much more refined.
There are some other updates, such as the battery capacity being increased by 50% compared to the previous generation, now allowing Figure 02 to work for 20 hours a day.
The wiring has also been redesigned, with integrated wiring for power and computing now using concealed wires, offering tighter packaging and higher reliability.
The exoskeleton structure of the body has been redesigned, balancing structural rigidity and impact load prevention. Of course, these changes have also increased Figure 02's weight to 70 kg, 10 kg heavier than 01.
The "World's Strongest" Robot?
Within 3 hours of Figure 02's release, it attracted the attention of 500,000 netizens.
Many expressed amazement: "Can't imagine what will happen in 20 years!" "2024 is absolutely the year of robots!"
Jim Fan, NVIDIA's senior scientist and head of embodied AI, immediately offered praise, stating: "The improvement in degrees of freedom of the fourth-generation hands is absolutely the right choice."
Like Jim Fan, many netizens were amazed by the smooth hand movements.
In fact, the birth of the fourth-generation hands stems from founder Adcock's persistence.
"We chose to make humanoid robots because the current world is built around human activities, with all standards adapted to human physiological conditions," Adcock once explained in an interview why he insists on making good humanoid robots.
Only by aligning everything with "humans" can we better serve people and help them save unnecessary labor.
His thinking aligns with most netizens - "The purpose of AI is not to write poetry or paint, but to wash dishes and do laundry for me, so I have time to write poetry and paint."
From deciding on the AI approach to becoming an industry leader, Adcock only took 2 years.
This AI company was only founded in 2022. Such rapid development relies on Adcock's foresight.
Before raising nearly $700 million for Figure AI, he had founded a software company and an aircraft company, with the sale of the former winning Adcock his "first bucket of gold". The latter has also successfully gone public.
With the arrival of the AI era, Adcock, like many others, decided to "All in AI". But unlike others, with the experience of two successful entrepreneurial ventures, the process of founding Figure AI was as smooth as if he had a golden touch.
As an undisputed "Silicon Valley nouveau riche", Figure AI is backed by joint investments from giants like Bezos (Amazon founder), OpenAI, NVIDIA, etc., with Figure AI currently valued at $2.6 billion (approximately 18.6 billion RMB).
Figure AI has lived up to expectations, with its product Figure 01 being the world's first commercially viable autonomous humanoid robot.
After 18 months, Figure 02 was officially released today. It is officially described as "the world's most advanced AI hardware".
However, where there are flowers, there are also doubts.
Some netizens posted demonstration videos of competitor Tesla's Optimus, stating that these improvements were already being made by Tesla 7 months ago, questioning how Figure 02 became the "most advanced".
The "Arch-rival" Optimus
In fact, as two of the most watched embodied AI projects in the tech world, the controversy between Figure AI and Tesla's Optimus has been ongoing.
A year ago, when Figure 01 released its teaser, some netizens joked: "Tesla's robot is called Optimus, so yours should be called Megatron."
During this release of Figure 02, some netizens eagerly expressed: "Can't wait to see Figure 02 battle Optimus Gen 2!"
Moreover, Adcock's own team includes many former Tesla employees.
Adcock didn't specifically respond to that questioning comment; he seems to have never cared about competition with Optimus.
Although Optimus has the "big tree" of Tesla behind it, with extensive data supply for training and ample research funds, Figure AI itself can be considered "well-provided for".
Not only does it have plenty of funds, but in terms of commercial cooperation, Figure AI is also "promising for the future". The collaboration with BMW is currently in its first phase, with Figure robots to be applied in the initial stages of car production. After the completion of the first phase, BMW will further collaborate deeply with them to jointly explore advanced technological themes such as artificial intelligence, robot control, manufacturing virtualization, and robot integration.
Optimus is currently also being used in Tesla factories. Both leading players in embodied AI are racing on their own paths.
As for who is the "world's most advanced AI hardware", this question need not be debated. The title won't disappear, but it will shift. In the rapidly developing AI industry, the next technology leader may already be waiting to take the stage.