Figure Humanoid

Figure AI's high-performance humanoid robots - Figure 01, 02, 03 generational evolution

Overview

Figure AI is a US robotics company founded in 2022 by Brett Adcock, focused on developing AI-powered general-purpose humanoid robots. Led by serial entrepreneur Adcock, who founded Archer Aviation and Vettery, Figure AI has rapidly emerged as a leader in the humanoid market. 1

ItemContent
CompanyFigure AI, Inc.
FounderBrett Adcock (CEO)
Founded2022
HeadquartersSunnyvale, California
VLA ModelHelix (Vision-Language-Action model)
Target MarketIndustrial (manufacturing, logistics), Home

Key Significance

Figure AI humanoids have significant importance in the Physical AI field:

  • First Commercial Humanoid Manufacturing Deployment (Figure AI claim): General-purpose humanoid deployed on actual production line at BMW factory 2
  • Vertical Integration AI Strategy: Transitioned from OpenAI partnership to in-house Helix VLA development (February 2025) 3
  • High-Performance Hardware-Software Integration: Combination of Helix VLA capable of 200Hz full-body control with optimized hardware 4
  • Aggressive Cost Reduction Target: Targeting 90% parts cost reduction in Figure 03, aiming for under $20,000 at mass production 5
  • Large-Scale Production Infrastructure: BotQ factory targeting 12,000 units annually (announced March 15, 2025) 6
  • Proven Industrial Performance: Contributed to production of over 30,000 X3s at BMW factory 2

Generational Comparison

ItemFigure 01 (2023)Figure 02 (2024)Figure 03 (2025)
PurposePrototypeIndustrial pilotCommercial mass production
Height168cm (5’6”)168cm (5’6”)168cm (5’6”)
Weight60kg (132 lbs)70kg (154 lbs)60kg (-14%)
Hand DoFBasic gripper16 DoF (both hands)Improved design
Total DoF24 DoF35 DoF35+ DoF
Hand Payload20kg20kg25kg
CamerasBasic6x RGB8x (including 2 palm)
BatteryIntegrated (5 hours)2.25 kWh (5 hours)2.3 kWh (5 hours)
ChargingWiredWiredWireless induction (2kW)
ComputingBasicNVIDIA RTX dualDual GPU (S1/S2)
ManufacturingHandmadeCNC machiningDie-casting/injection molding
Target Price-~$30K (estimated)Under $20K (target)

Source: Wikipedia, Figure AI official announcements


Figure 01

Figure AI’s first humanoid robot officially announced on March 2, 2023. Took its first steps in May 2023. 1

Physical Specifications

ItemSpec
Height168cm (5’6”)
Weight60kg (132 lbs)
PayloadUp to 20kg (44 lbs)
Degrees of Freedom24
Operating Frequency200 Hz
Operating TimeUp to 5 hours
Walking Speed1.2 m/s

Key Features

  • Bipedal Robot: Targeting logistics and warehouse operations
  • Basic Manipulation Capabilities: Climbing stairs, lifting boxes, using tools
  • OpenAI Integration: Voice conversation and reasoning capabilities using large language models
  • Human-Level Dexterity: Capable of performing tasks requiring precision and coordination
  • Torque-Controlled Walking: Adapts to uneven terrain and external disturbances

Figure 02

2nd generation humanoid announced August 6, 2024, marking the serious stage for industrial deployment. 1

Physical Specifications

ItemSpec
Height168cm (5’6”)
Weight70kg (154 lbs)
PayloadUp to 20kg
Walking Speed1.2 m/s
Battery2.25 kWh (custom)
Operating TimeOver 5 hours

Key Features

ItemSpec
Hand DoF16 (5-finger both hands)
Total DoF35
Hand PayloadUp to 25kg
TorqueUp to 150Nm
Range of MotionUp to 195 degrees
Cameras6x RGB
ComputingDual NVIDIA RTX GPU (3x previous)
SensorsRGB cameras, IMU (Inertial Measurement Unit), gyroscope, force sensors, non-contact sensing, microphones, speakers

Major Improvements

  • Cable integrated design in limbs (sleek matte black exterior)
  • Battery integrated in torso (50% more energy than previous)
  • Onboard VLM (Vision Language Model)
  • Real-time perception, decision-making, execution capable
  • Speech-to-speech conversation via OpenAI models

Figure 03

3rd generation humanoid announced October 9, 2025, mass production model targeting both home and commercial use. Initial partner deployments began the same day. 5

Physical Specifications

ItemSpec
Height168cm (5’6”)
Weight60kg (Figure 02 -14%)
Battery2.3 kWh (custom, 78% cost reduction vs Figure 02)
Operating TimeAbout 5 hours (300 minutes)
ChargingWireless induction 2kW (foot coils)
Data Transfer10 Gbps mmWave

Sensor System

  • Cameras: 8 (6 main + 2 palm)
  • Tactile Sensors: Custom sensors on fingertips (sensing down to 3g - paper clip weight)
  • Vision System: 2x frame rate, 1/4 latency, 60% wider FOV

Audio System

  • Speakers: 2x size, 4x output
  • Microphones: Repositioned for improved clarity
  • Real-time speech-to-speech conversation support

Safety Features

  • Multi-density foam for pinch point protection
  • Soft textile exterior (instead of hard metal)
  • UN38.3 certified battery (BMS, cell, interconnect, pack level protection)

Manufacturing Innovation

  • Manufacturing Method Transition: CNC machining -> die-casting, injection molding, stamping
  • Parts Cost Reduction Target: 90% reduction
  • Target Price: Under $20,000 (at mass production)
  • Wrist Redesign: Lessons learned from BMW deployment reflected, distribution board and dynamic cabling removed

Helix VLA

Helix is Figure AI’s in-house developed VLA (Vision-Language-Action) model announced February 2025. Figure AI introduces Helix as “the first VLA for high-speed continuous control of a humanoid’s full body.” 4

Architecture

System 1 / System 2 Dual System Structure:

SystemRoleFrequencyParameters
System 2 (S2)High-level planning, VLM7-9 Hz7B
System 1 (S1)Low-level control, real-time200 Hz80M

Key Features

  • 35 DoF Control: Full upper body control including fingers, wrists, torso, head
  • Figure AI’s Claimed Firsts:
    • First VLA for high-speed continuous humanoid full-body control
    • First VLA for dual robot simultaneous control
    • First VLA running fully onboard on embedded low-power GPU
  • Learning Efficiency: Trained on approximately 500 hours of teleoperation demos
  • Versatility: No task-specific adaptation needed, single weights for various tasks

Representative Performance: Table-to-Dishwasher

ItemValue
Distance Traveled130+ feet
Unique Interactions33
Objects Handled21 (including delicate dishware)

According to Figure AI, evaluated as “the most complex task ever performed autonomously by a robot.” 4


Industrial Deployment

BMW Factory Pilot (2024-2025)

According to Figure AI, this is the first commercial deployment of a general-purpose humanoid in an automotive production facility. 2 7

ItemContent
LocationBMW Spartanburg Plant (South Carolina)
Duration11 months
Robots Deployed2 Figure 02 units
Work Hours5 days/week, 10-hour shifts
Total Operating Hours1,250 hours

Results

  • Parts Processed: Over 90,000 sheet metal parts loaded
  • Production Contribution: Contributed to over 30,000 BMW X3 production
  • Task Description: Pick sheet metal parts from racks/bins and place on welding equipment
  • Total Distance Traveled: Approximately 200 miles (322km)
  • Accuracy: Over 99%

Performance Requirements

ItemStandard
Placement AccuracyWithin 5mm tolerance
Single Motion Time2 seconds
Part Loading Time37 seconds
Full Cycle Time84 seconds
Success Rate Target99% per shift

Lessons Learned

Key lessons from BMW deployment reflected in Figure 03 design:

  • Forearm Issues: Most frequent hardware failure point -> Wrist electronics completely redesigned in Figure 03
  • Thermal Management: Issues due to tight packaging and agility requirements -> Distribution board and dynamic cabling removed

Funding and Valuation

Investment History

PeriodRoundAmountValuationKey Investors
2023.05Series A~$70M-Led by Parkway Venture Capital
2024.02Series B$675M$2.6BJeff Bezos, Microsoft, NVIDIA, Intel Capital, Amazon Industrial Innovation Fund, OpenAI Startup Fund, ARK Invest, Align Ventures
2025.09Series C$1B+$39BLed by Parkway Venture Capital, Brookfield, NVIDIA, Macquarie Capital, Intel Capital, Align Ventures, LG Technology Ventures, Salesforce, T-Mobile Ventures, Qualcomm Ventures

Source: Wikipedia, PRNewswire

Cumulative Funding: Approximately $2B (Series A + B + C)

OpenAI Partnership

  • February 2024: Collaboration agreement with OpenAI, joint development of next-generation AI models for humanoids 8
  • February 2025: Partnership ended, transitioned to in-house AI development 3
  • Reason: “To solve Embodied AI at scale in the real world, you need to vertically integrate robot AI” - Brett Adcock

Future Plans

  • BotQ Factory: Announced March 15, 2025, targeting 12,000 units annually as the largest humanoid factory in the US 6
  • 100,000 units within 4 years plan (roadmap)
  • Home Market Entry: Consumer market entry projected for 2026, Figure 03’s household helper role (laundry, cleaning, dishwasher, etc.) 5
  • Figure 02 Retirement: Fleet-wide retirement of Figure 02 underway following Figure 03 release

Glossary

TermDescription
VLAVision-Language-Action model. AI model integrating visual input, language understanding, and action output
VLMVision-Language Model. AI model that understands images and text together
DoFDegrees of Freedom. Number of independent directions a robot can move
IMUInertial Measurement Unit. Sensor measuring acceleration and angular velocity
FOVField of View. Range a camera can see

References

Additional References:


See Also

Footnotes

  1. Figure AI Wikipedia 2 3

  2. Figure AI - BMW Production Contribution 2 3

  3. Figure drops OpenAI - TechCrunch 2

  4. Figure AI - Helix Announcement 2 3

  5. Figure 03 Announcement 2 3

  6. BotQ Announcement 2

  7. BMW Official Press Release

  8. Series B Funding Announcement