On October 9th, the well-known American humanoid robot startup Figure AI officially launched its third-generation humanoid robot “Figure 03” on its official website. To showcase the new product’s capabilities, the company released a 6-minute and 19-second introduction video, which intuitively demonstrates Figure 03’s versatility across both home and commercial scenarios. In the footage, the robot can smoothly carry out household tasks such as serving tea and water, playing with pets, tidying rooms, washing dishes, and folding clothes. It is also competent in commercial roles, including hotel front desk service, express delivery, package sorting, and banquet waiting. Compared to its predecessor, Figure 03 boasts significantly improved movement smoothness, breaking the movement limitations that have long constrained traditional robots in complex environments.
In terms of appearance design and material selection, Figure 03 breaks away from the traditional metal structure framework, and the overall outer layer is covered with fabric. Figure AI said that this soft covering layer can not only enhance the safety and affinity when the robot interacts with humans, but also has the characteristics of being easy to disassemble, replace, and clean. Users can match the robot with a variety of clothes according to their needs, further adapting to the visual needs of different scenarios and allowing the robot to blend into the home or commercial environment more naturally.
The hand structure is one of the core upgrade highlights of Figure 03. Its fingers are made of a softer and more compliant material, and the contact area is greatly increased, which can stably grasp objects of different shapes and sizes. More precisely, each fingertip can accurately sense a pressure of only 3 grams – this weight is equivalent to that of a paper clip, which is enough to distinguish the state of “stable grip” from “about to slip”, providing precise control for fine operations. In a six-second close-up video, Figure 03 gently picked up a tomato, an egg, and a playing card respectively through different gestures and put them down smoothly. Especially when picking up the top one of a stack of playing cards, the fineness and coordination of its hand movements were vividly demonstrated. In addition, each palm is also equipped with a wide-angle, low-latency camera, so that even if the main camera is blocked by cabinets, narrow spaces, etc., it can provide close-range visual feedback, ensuring the continuity and real-time performance of the operation.
The optimization of the visual and audio systems has further enhanced the environmental adaptation and interaction capabilities of Figure 03— a key highlight in the latest robotic news. After the visual architecture has been comprehensively upgraded, the frame rate has doubled compared to the previous generation, the latency has been reduced by a quarter, the single-lens field of view has been expanded by 60%, and it also has a wider depth of field range. These improvements provide a more intensive and stable perception input for Figure AI’s exclusive “Helix Visual-Language-Action Artificial Intelligence Model”, helping the robot to achieve more precise intelligent navigation and operation in complex environments such as the home. In terms of audio hardware, the speaker size of Figure 03 has doubled compared to the previous generation, the power has been increased to nearly four times, and the position of the microphone has also been readjusted, which greatly optimizes the real-time performance and clarity of voice conversion, making the communication between humans and the robot more natural and smooth.
In the aspect of practicality and mass production adaptation, Figure 03 has also been designed specifically. To adapt to the movement needs of the home space, its body mass has been reduced by 9% compared to Figure 02, and the volume has been significantly reduced, making it easier to move through narrow areas. The battery system has passed the UN38.3 international safety certification and added multiple protection mechanisms to ensure safe use; the bottom of the foot is equipped with an induction coil, which can be charged at a power of 2kW through a wireless base, and the robot can also autonomously return to the charging position to recharge and upload data. In addition, Figure 03 supports 10Gbps millimeter-wave data transmission, which can transmit learning data back at high speed to achieve continuous iterative optimization, becoming the first humanoid platform to truly achieve end-to-end learning “from pixels to actions”.
For the commercial scenario, the speed of the new driver of Figure 03 has been increased to twice, and the torque density has been significantly improved, which can complete the tasks of transporting and placing items more efficiently. It is worth noting that this robot has been designed with the goal of “high-volume manufacturing” from the initial stage of design. Almost all parts have been redesigned, and by simplifying the assembly steps and reducing the number of parts, the production cost has been effectively reduced, laying the foundation for subsequent large-scale applications.
