Brett Adcock, CEO and founder of Figure, provides an exclusive tour of the company's robot campus in Sunnyvale, where they design, manufacture, and test humanoid robots. Figure employs over 500 people across four buildings, with 250-300 working on-site developing what Adcock claims is the world's most advanced general-purpose humanoid hardware.

The conversation covers Figure's complete robotics pipeline, from their proprietary Helix neural network that controls all robot movements to their manufacturing facility called 'Bot HQ' where robots are assembled and tested. Adcock discusses the technical challenges of creating robots with human-level stability and dexterity, the company's rapid progression through three robot generations, and their vision for deploying robots in both commercial and residential settings.

Drawing inspiration from science fiction, particularly Contact which Adcock cites as his favorite, Figure aims to achieve artificial general intelligence through embodied AI that learns from real-world physical interaction rather than purely digital training.

Neural Network Control System Powers 40-Joint Humanoid

Figure's robots contain 40+ motors that can each rotate 360 degrees, creating 360^40 possible body positions - mathematically more combinations than atoms in the universe.

The proprietary Helix neural network runs onboard at 50-200 milliseconds intervals, controlling every joint position from camera input without requiring internet connectivity.

"You can't write code to make this work. All of our robots here run on a neural network we call Helix" - Brett, explaining the transition from traditional robotics programming to AI-first control.

Robots operate fully autonomously for 4-5 hours on battery, then automatically dock and wirelessly charge through their feet at 2 kilowatts for one hour.

Fault Tolerance Through Reinforcement Learning

Figure developed 'Vulcan' technology allowing robots to continue operating after losing critical joints like knees or ankles, preventing falls through neural network adaptation.

"We have an initiative internally called never fall. We never ever want to fall" - Brett, describing their zero-tolerance approach to robot stability failures.

The fault tolerance system learns in simulation to handle joint failures, then transfers zero-shot to physical robots without additional training.

When a knee joint fails and locks, robots can 'hobble around' and continue work or request replacement robots to maintain 24/7 operations.

Manufacturing Scale and Cost Reduction

Figure reduced per-robot manufacturing costs by 90% between generations 2 and 3, dropping from hundreds of thousands to under $100,000 each.

The company achieved record manufacturing in March 2024, producing more robots than their entire previous company history combined.

Manufacturing facility 'Bot HQ' handles complete assembly from heads and batteries to final testing, with robots walking directly to headquarters upon completion.

Figure plans full 'lights out' manufacturing where robots will build, test, and box themselves for shipment without human intervention.

Commercial Deployment and BMW Partnership

Figure deployed robots at BMW's factory for six months, helping build what they claim is the first car in the world assembled by humanoid robots.

"This was the first car in the world built by a humanoid robot that we're aware of" - Brett, showing the BMW X3 they helped manufacture and purchased.

Robots successfully operated on BMW's body shop assembly line, demonstrating commercial viability for manufacturing applications.

Figure's robot was invited to the White House as the first humanoid robot in history to operate there, greeting visitors and demonstrating capabilities.

Home Deployment Strategy and Pricing

Figure plans to lease home robots for $400-600 per month, similar to car lease pricing, rather than selling units directly to consumers.

Home robots require only a 2x2 foot charging dock that plugs into standard wall outlets, with robots automatically docking when battery reaches 10-15%.

Target applications include daily laundry, dishes, and house tidying, with robots operating the same Helix neural network used for commercial applications.

Data collection from deployed robots will be anonymized and used for central training to improve the global fleet through software updates.

Training Data and AI Development

Helix neural networks train on nearly one million hours of base data, with additional thousands of hours in post-training for specific applications.

"The biggest blocker for us now of going from where we're at today to large scale deployment is data" - Brett, identifying data collection as the primary scaling challenge.

Figure employs motion capture specialists in full-body suits to collect human movement data for training humanoid behaviors and dexterity.

The company believes embodied AI learning through physical world interaction may achieve artificial general intelligence faster than purely digital approaches.

Design Evolution and Future Generations

Figure progressed from expensive CNC-manufactured Figure 1 to mass-producible Figure 3, with Figure 4 representing what Adcock calls their 'iPhone moment.'

"Figure four will be the biggest step up we've ever made by far... it's just like radically different" - Brett, describing the upcoming generation as transformational.

Hand technology evolved from complex tendon-driven systems to direct motor control, with next-generation hands achieving human-level dexterity and joint count.

Design philosophy balances functionality with approachability, choosing 'Westworld' humanlike aesthetics over 'I, Robot' mechanical appearance, inspired by sci-fi including Contact.