news

Figure AI created the "Terminator"

2024-08-07

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Machine Heart Report

Synced Editorial Department

Capable of voice conversation, VLM vision, and working 20 hours a day.

This day will always come, but we didn't expect it to come so soon.

On the evening of August 6th, Beijing time, Figure, a well-known Silicon Valley embodied intelligence startup, officially released its new generation of humanoid robot Figure 02.





In addition to looking sci-fi, this robot also has enough general intelligence to communicate with humans in real time and automatically learn how to assemble parts. In fact, Figure 02 is currently interning at BMW's factory in Spartanburg. It seems like we have already entered the future.



The Figure engineering and design team completed a complete redesign of the Figure 02 hardware and software. Significant advances were made in key technologies including artificial intelligence, vision, batteries, electronics, sensors, and actuators.

Synced

,Thumbs up

539

Specifically, Figure 02 mainly includes the following features:

  • Real-time voice conversation:Figure 02 Able to converse with people through built-in microphone and speaker connected to custom OpenAI big model;
  • Camera: AI vision system driven by 6 RGB cameras;
  • Robot: A fourth-generation manipulator with 16 degrees of freedom and human-like strength;
  • Built-in VLM: Supports fast common-sense visual reasoning for robot cameras;
  • Large capacity battery: A 2.25 KWh custom battery pack in the robot’s torso provides 50% more power;
  • CPU/GPU: Onboard computing and AI reasoning capabilities have increased by 3 times compared to the previous generation.

Comprehensive improvement: general + humanoid + practical

The biggest change with this upgrade may be that Figure 02 is actually ready to respond to voice commands.

In the demo that amazed the tech world last year, a human standing in front of Figure 01 asked the robot, "Can you give me something to eat?" It not only recognized that the object in front of it was an apple, but also understood that apples can be eaten directly, so it handed the apple to the person in front of it and said, "Of course, take it and eat it."

In March this year, OpenAI announced a collaboration with Figure to develop embodied intelligence, enabling humanoid robots to have real-time and effective conversational capabilities. Backed by OpenAI, Figure can quickly iterate the robot's voice-to-voice capabilities. Figure said that due to the significant improvement in the end-side computing power of Figure 02, it can now perform various tasks in the real world completely autonomously.

In Figure 02, the robot’s voice reasoning is made possible by the built-in microphone and speaker connected to a custom AI model trained in collaboration with OpenAI.



Figure 02’s navigation system uses VLM (Visual Language Model), which enables the camera on the robot to perform semantic basic research and fast common-sense visual reasoning. VLM is a new direction of intelligent driving that many new car manufacturers are currently studying. In the field of robotics, this method can obviously achieve breakthrough results in many tasks.



Battery life is one of the biggest challenges facing the practical application of humanoid robots. The Figure 02 robot has a custom 2.25 KWh battery pack in its torso, which can provide more than 50% more energy than the Figure 01, thereby maximizing the robot's operating time. Brett Adcock, founder and CEO of Figure AI, said they hope that Figure 02 can achieve more than 20 hours of effective work per day.



We can see that the appearance of Figure 02 has become more integrated because it uses a new exoskeleton structure. Compared with Figure 01, the robot's appearance design has changed more thoroughly. In addition, it also uses an integrated wiring design. Integrated wiring has the following benefits:

  • Higher reliability
  • Hide the wires
  • Tighter packaging



From Figure 01 to Figure 02, there is a significant change in appearance.

In addition to better wiring, Figure 02 also designed custom wire terminals and connectors to improve the robot’s reliability.



In order to understand the world like humans, Figure 02 uses a multi-camera + AI-driven visual system for perception and reasoning. It has a total of 6 onboard RGB cameras on the head, front torso and back torso, giving the robot visual capabilities that surpass humans.



Finally, there are the nimble fingers. Figure 02 is equipped with a fourth-generation manipulator with 16 degrees of freedom and human-like strength, which can handle a variety of complex tasks. The entire hand includes mechanical, electrical, control and sensor technologies.



Brett Adcock said that as the robot continues to run, the AI ​​data engine will collect and organize data for training models and continuously improve performance.



Seeing Figure 02’s wonderful appearance, people can’t help but think of Musk’s Optimus Prime. Comparing the two, I wonder which one is better.



Another netizen said, "Figure 02 heralds the beginning of a new era. It is the most advanced robot in the world. The future has arrived."



Figure, a startup: Half of Silicon Valley is investing in

Founded in 2022 by Brett Adcock, Figure is a US-based robotics company that specializes in developing humanoid robots. The company's goal is to develop universal humanoid robots that will have a positive impact on humanity and create a better life for future generations.

After the outbreak of generative AI technology, many robotics companies aiming at embodied intelligence have emerged, and Figure stands out among them. In March 2023, Figure came out of stealth mode and launched a prototype robot, Figure 01, which looks and moves like a human. This is a bipedal robot designed for manual labor, initially targeting the logistics and warehousing industries.

In May 2023, the company raised $70 million from investors led by Parkway Venture Capital.

In October of the same year, Figure released a video of the Figure 01 bipedal robot walking.



Fast forward to January of this year, and Figure 01 had learned how to make coffee.



On January 18, 2024, Figure announced a partnership with BMW to deploy humanoid robots in car manufacturing plants. At this time, Figure 01 is said to be able to complete tasks in the real world autonomously.



Figure 01 Working in a BMW factory.

In March this year, Figure announced the completion of a staggering $675 million Series B financing, with the company's valuation reaching $2.6 billion. Investors include Microsoft, Intel, OpenAI Startup Fund, Amazon Industrial Innovation Fund, Nvidia, Bezos, Cathie Wood's Ark Investment, Parkway Venture Capital, Align Ventures, etc.

At the same time, Figure also announced a partnership with OpenAI, which includes OpenAI building a specialized AI model for Figure's humanoid robot, enabling its robots to process and reason about language.

On March 13, with the help of OpenAI technology, Figure 01 can have a full conversation with humans.

From walking upright, completing complex tasks to interacting naturally with people. These technical highlights are also one of the important reasons why Figure and OpenAI, which has always wanted to return to the field of robotics, reached a cooperation agreement - combining OpenAI's research with Figure's robotics experience to develop the next generation of AI models for humanoid robots.

With today’s release of Figure 02, the combination of highly integrated hardware and next-generation AI technologies like VLM brings us one step closer to truly humanoid robots with general capabilities.

Soon, the physical world will also usher in changes brought about by AI?

References:

https://x.com/Figure_robot/status/1820791819023909031

https://www.youtube.com/watch?v=0SRVJaOg9Co

https://www.therobotreport.com/figure-02-humanoid-robot-is-ready-to-get-to-work/