How Is Figure Building Humanoid Robots at Scale?

TL;DR

Figure is developing general-purpose humanoid robots through rapid hardware iteration, with Figure 2 serving as its feature-complete second-generation platform. The robot adds substantially more onboard computing, a larger battery, six cameras, internal wiring, load-bearing exterior shells, and improved hands, while future generations will focus on reducing costs by more than tenfold and enabling large-scale manufacturing.

Transcript

I am curious how you think about Optimus and Tesla and Elon so inspired by what elon's done in the last like 20 30 years it's been just unbelievable we need more like really real players I think Tesla is a really real player like I mean you know highly capitalized great engineering team and I think they're directly heading in the right direction I'... Read More

Key Insights

• Figure 2 is a second-generation humanoid robot manufactured at Figure's Northern California facility. At the time of the interview, the company had more than 100 engineers, fewer than 10 robots operating internally, and a production rate of approximately one robot per week.
• Figure 2 has three times the onboard CPU and GPU capacity of Figure 1. The additional computing power is intended to support more inference directly on the robot, strengthening its ability to process information and operate through its onboard systems.
• Figure 2 uses a battery with approximately 2.3 kilowatt-hours of capacity. The battery is integrated into the middle of the torso near the computing hardware, and its capacity is nearly twice that of the first-generation robot.
• Figure 2 uses six onboard cameras to perceive its surroundings. The cameras are positioned in the head, back, and lower torso, giving the robot multiple viewpoints for sensing its environment and supporting perception during operation.
• Figure 2 has internal wiring, cabling, and electronics rather than exposed connections. This packaging choice is intended to improve reliability and create a more integrated physical system that is better suited to continued testing and eventual commercial use.
• Figure 2 uses an exoskeleton structure in which the outer shells carry loads. Combining structural stiffness and crash-load management within the same components helps avoid the redundant mass created when a robot has both a separate internal structure and exterior shell.
• Figure's fourth-generation robotic hands improve sensing, packaging, mass, strength, finger speed, dexterity, and fine-grained manipulation. These capabilities are necessary because a general-purpose humanoid must perform humanlike applications and grasp objects designed for human hands.
• Figure's development strategy separates hardware maturity into successive generations. Figure 1 addressed approximately 100 to 200 architectural decisions, Figure 2 targeted feature completeness, and later versions will focus on reducing cost by well over tenfold and enabling manufacturing at unprecedented scale.

Questions & Answers

Q: What are the main upgrades in the Figure 2 humanoid robot?

Figure 2 triples onboard CPU and GPU capacity, nearly doubles battery capacity to approximately 2.3 kilowatt-hours, and incorporates six cameras positioned around the head, back, and lower torso. It also moves wiring, electronics, and cables inside the body, adopts load-bearing exterior shells, and uses fourth-generation hands with improved sensors, packaging, strength, finger speed, dexterity, and control.

Q: How does Figure 2 perceive its surroundings?

Figure 2 perceives its environment through six onboard cameras. These cameras are distributed across the head, back, and lower torso, providing multiple views of the robot's surroundings. The expanded perception system is paired with three times more onboard CPU and GPU capacity, giving the robot additional computing resources for inference and processing information collected by its cameras.

Q: Why does Figure 2 use a load-bearing exoskeleton?

Figure 2 uses exterior shells that carry structural loads, combining stiffness and crash-load requirements within the same components. Figure 1 used both an underlying structure and an outer shell, which could create redundant mass because the internal structure had to be sized for crash loads. The exoskeleton approach is intended to produce a stiffer, lighter, and more efficiently packaged robot.

Q: How are Figure 2's hands designed for humanlike tasks?

Figure 2 uses Figure's fourth-generation robotic hands, which improve sensors, packaging, mass, strength, finger speed, dexterity, and fine-grained control. These improvements are intended to help the robot perform humanlike applications and grasp objects made for human use. Better manipulation also supports Figure's broader goal of creating a general-purpose robot instead of a machine limited to one specialized task.

Q: How quickly was Figure producing humanoid robots?

At the time of the recorded conversation, Figure was manufacturing approximately one humanoid robot per week at its Northern California facility. The company had fewer than 10 robots operating internally and employed more than 100 engineers. This production level supported ongoing testing while Figure continued refining its second-generation platform and preparing for later efforts focused on cost reduction and much larger manufacturing volumes.

Q: Why does Figure rapidly create new hardware generations?

Figure's development strategy assumes that a minimum of roughly three hardware versions may be necessary before the platform becomes commercially reliable, safe, low-mass, low-cost, and easy to manufacture. The company uses each generation to mature different parts of the system, instead of expecting its first robot to resolve every architectural, mechanical, electrical, firmware, control, and manufacturing decision simultaneously.

Q: What roles do Figure 1 and Figure 2 play in development?

Figure 1 primarily established the robot's architecture by addressing approximately 100 to 200 decisions across systems such as batteries, actuators, electronics, wiring, sensors, software, and controls. Figure 2 focused on becoming feature complete, meaning the planned hardware and software systems were present and working. Future generations are expected to concentrate more heavily on affordability and scalable manufacturing.

Q: How does Figure plan to reduce humanoid robot costs?

Figure plans to analyze the robot's complete bill of materials from the bottom up, itemizing approximately 1,000 parts and determining how each component should be procured at scale. The company evaluates whether to build or buy each part and examines contractual volume pricing. Its future-generation objective is to lower current costs by well over tenfold while enabling unprecedented manufacturing scale in robotics.

Summary & Key Takeaways

• Figure 2 is Figure's second-generation humanoid robot, developed by a company roughly two years old at the time of recording. Figure employed more than 100 engineers in Northern California, operated fewer than 10 robots at its facility, and was manufacturing approximately one robot per week when the conversation occurred.

• The new robot triples onboard CPU and GPU capacity, nearly doubles battery capacity to approximately 2.3 kilowatt-hours, and includes six cameras. Internal wiring improves packaging and reliability, while load-bearing exterior shells reduce redundant structural mass. Fourth-generation hands add better sensing, speed, strength, packaging, dexterity, and fine manipulation.

• Figure's strategy assumes at least three hardware versions may be required to reach commercially reliable hardware. Figure 1 established major architectural choices, while Figure 2 became a feature-complete system. Future development will emphasize reducing costs by well over tenfold and creating manufacturing processes capable of unprecedented scale in robotics.