Figure's latest humanoid robot: more dexterous hands, more reliable for factory screwdriving!

08/08 2024 389

On August 7, Beijing time, Figure, an American humanoid robotics startup, unveiled its second-generation humanoid robot, Figure 01. Compared to its predecessor, Figure 02, as seen in the released video, boasts flexible finger movement, faster walking speed, longer battery life, and the integration of a Vision-Language Model (VLM).

According to Brett Adcock, Founder and CEO of Figure, Figure 02's battery capacity has increased by 50%, enabling it to support 20 hours of effective work per day. Starting next year, Figure plans to facilitate low-cost, high-efficiency mass production, thereby entering the labor market and home application market. Currently, Figure 02 is already in use for screwdriving at BMW factories.

A More Flexible Figure 02

Flexibility is one of the key objectives pursued by all humanoid robots, professionally referred to as degrees of freedom (DOF).

To achieve greater body flexibility, Figure 02 adopts a fourth-generation hand design featuring 16 DOF, which imparts remarkable hand dexterity to Figure 02, significantly enhancing the precision of its operations.

Meanwhile, the highly anticipated Tesla Optimus Gen2, the second-generation humanoid robot, boasts only 11 DOF in its hands, while most other companies offer 6-7 DOF.

The higher the DOF, the greater the robot's flexibility and mobility.

Apart from increased DOF, Figure 02's hand features a dot matrix black skin that potentially integrates tactile sensors, endowing it with tactile perception and the friction necessary for grasping and manipulating various objects.

As seen in the following video, Figure 02's fingers exhibit remarkable flexibility, and its limbs, arms, head, hands, and fingers move with agility, demonstrating proficient assembly of automotive parts in BMW factories.

Of course, hand dexterity alone is insufficient; joint flexibility is equally crucial. To enhance flexibility, Figure 02 underwent hardware and software iterations. Its robotic arm employs six-dimensional sensors, crucial for simulating joint and limb-end tactile sensations. Precise speed and torque adjustments via reducers enable precise robot control.

Increasing flexibility necessitates advancements in core perception and motion algorithms, though limited information has been disclosed thus far.

A Smarter Figure 02

Figure 02's intelligence stems from its integration of a Vision-Language Model (VLM), enabling more natural speech interaction with humans.

On the hardware front, Figure 02 boasts six cameras, a more powerful processor that triples inference speed with the support of large models, and the addition of speakers and microphones, effectively granting it a mouth and ears.

One conceivable work scenario involves Figure 02 engaging in voice conversations with fellow human workers on the factory line while performing operations with its agile body.

The combination of OpenAI's large models and Figure 02 can propel the development of embodied intelligence. This represents the true integration of hardware and software, the direction of AI development, enabling AI to perform physical tasks like screwing and household chores rather than indulging in poetry and distant dreams.

Numerous global companies are attempting to integrate large models with humanoid robots. Eventually, Grok, the large model from Musk's xAI, will likely converge with the humanoid robot Optimus.

To prove their worth, humanoid robots must first excel at factory screwdriving. Some, like Unitree Go 2 Pro, have even seen combat duty, with a reported price of 18,600 yuan.

This video, shared by a verified military blogger on Weibo with millions of followers, showcases the cost-effectiveness of adding combat modules to such robots. Figure has already been tested in BMW's Spartanburg, South Carolina, factory, while Tesla has tested Optimus in its own factories. Similarly, Hyundai Motor Group's Boston Dynamics, Agility, Apptronik, and Sanctuary AI have announced similar tests with automakers.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.