Figure AI
Last reviewed
Sources
53 citations
Review status
Source-backed
Revision
v14 · 10,390 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
53 citations
Review status
Source-backed
Revision
v14 · 10,390 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Figure AI | |
|---|---|
| Type | Private company |
| Industry | Robotics, artificial intelligence, humanoid robots |
| Founded | May 2022 |
| Founder | Brett Adcock (Founder and CEO) |
| Headquarters | Sunnyvale, California, US |
| Products | Figure 01, Figure 02, Figure 03 humanoid robots; Helix AI system |
| Funding | Over $1 billion Series C at $39 billion post-money valuation (September 2025); ~$1.85 billion total raised |
| Key investors | Microsoft, NVIDIA, OpenAI Startup Fund, Jeff Bezos, Intel Capital, Parkway Venture Capital, Brookfield Asset Management |
| Website | figure.ai |
Figure AI is an American artificial intelligence and robotics company, founded in May 2022 by Brett Adcock, that designs and manufactures general-purpose humanoid robots powered by its own in-house AI. It is one of the most valuable private robotics companies in the world, reaching a $39 billion post-money valuation after raising more than $1 billion in a Series C round announced in September 2025.[1][40] The company has shipped three robot generations (Figure 01, Figure 02, and Figure 03), runs them on a proprietary vision-language-action model called Helix, and has deployed humanoids commercially at a BMW automotive plant where a single Figure 02 contributed to building more than 30,000 vehicles.[6][9]
Figure AI, Inc. (commonly referred to as Figure) is an American artificial intelligence and robotics company developing general-purpose humanoid robots. Founded in May 2022 by serial entrepreneur Brett Adcock, the company is headquartered in Sunnyvale, California. Figure has produced three generations of humanoid robots: the Figure 01, Figure 02, and Figure 03. The company's robots are powered by Helix, a proprietary vision-language-action model (VLA) developed in-house that enables autonomous manipulation, locomotion, and language understanding.
As of October 2025, Figure has raised over $1.85 billion in total funding and reached a post-money valuation of $39.5 billion, making it one of the most highly valued private robotics companies in the world.[40] The company's investors include Microsoft, NVIDIA, OpenAI, Jeff Bezos, Intel Capital, and Brookfield Asset Management. Figure has deployed its robots commercially at BMW manufacturing plants and in logistics warehouse operations. In April and May 2026, the company disclosed that its BotQ facility had delivered more than 350 Figure 03 units at a rate of one robot per hour, and demonstrated multi-day continuous autonomous operation, including an 81-hour run in which a single robot sorted more than 101,000 packages without human intervention.[55][59]
| Detail | Information |
|---|---|
| Type | Private |
| Industry | Robotics, artificial intelligence, humanoid robots |
| Founded | May 2022 |
| Founder and CEO | Brett Adcock |
| Headquarters | Sunnyvale, California, United States |
| Key people | Brett Adcock (CEO and Founder), Jerry Pratt (former CTO), Dana Votypka (CFO), Bob Klunk (COO), Lee Randaccio (VP of Growth) |
| Employees | ~400-500 (early 2026) |
| Total funding | ~$1.85 billion |
| Valuation | $39.5 billion (October 2025) |
| Products | Figure 01, Figure 02, Figure 03, Helix AI System |
| Website | figure.ai |
Brett Adcock founded Figure AI in May 2022 after leaving Archer Aviation, the electric vertical takeoff and landing (eVTOL) company he had co-founded in 2018. Before Archer, Adcock co-founded Vettery, an AI-powered recruiting platform that the Adecco Group acquired for approximately $110 million in 2018. He also took Archer public on the New York Stock Exchange in February 2021 at a $2.7 billion valuation.
Adcock self-funded Figure's initial operations with approximately $100 million of his own capital. He assembled the founding team by recruiting engineers from Boston Dynamics, Tesla, Google DeepMind, Apple, Agility Robotics, Rivian, and the Institute for Human & Machine Cognition (IHMC), bringing together over 100 years of combined experience in AI and humanoid robotics. Jerry Pratt, who had spent 20 years as a senior research scientist at IHMC leading bipedal locomotion research (his team placed second at the DARPA Robotics Challenge Finals) and held a PhD from MIT in computer science, joined as founding CTO. Pratt later departed Figure in 2025 and co-founded Persona AI with Nic Radford, another veteran roboticist.
Other early team members included Dana Berlin (VP of Commercialization and Capital), David McCall (Principal Industrial Designer, previously at Rivian, Audi, and Ford), Mathew DeDonato (Director of Robotic Systems, previously at Toyota/Woven Planet), and Rob Gruendel (Robotics Safety Lead), who sits on ANSI/ISO robotics safety standards committees.
The company operated in stealth mode for its first year, focusing on rapid prototype development. In 2022, Figure introduced Figure 01, a bipedal robot designed for manual labor in logistics and warehousing sectors. The robot achieved dynamic bipedal walking within 12 months of the company's inception, marking one of the fastest development timelines in humanoid robotics history.
Figure emerged from stealth in March 2023, simultaneously announcing its first humanoid robot, Figure 01, and completing initial funding. The company released CGI renders of the design at that time. By October 2023, Figure began publishing videos demonstrating autonomous capabilities, including dynamic bipedal walking. By late 2023, the company employed approximately 80 people and was rapidly scaling its engineering team.
In May 2023, Figure closed a $70 million Series A funding round led by Parkway Venture Capital, establishing a valuation of approximately $500 million.[3] Participating investors included Bold Capital Partners, Aliya Capital, Tamarack Global, FJ Labs, and Till Reuter (CEO of KUKA Robotics). An extension in July 2023 added $9 million from Intel Capital at a $350 million valuation, strengthening the company's balance sheet and enabling it to build out a data pipeline for autonomous operations.
On January 18, 2024, Figure signed a commercial agreement with BMW Manufacturing to deploy general-purpose humanoid robots in automotive production environments. This marked Figure's first major commercial partnership and the first commercial deployment of humanoid robots in automotive manufacturing.
In February 2024, the company announced a $675 million Series B round at a $2.6 billion valuation, representing a roughly fivefold increase from the Series A valuation less than a year earlier.[4] The investor consortium attracted high-profile strategic and venture backers:
| Investor | Type |
|---|---|
| Microsoft | Strategic (Azure infrastructure) |
| OpenAI Startup Fund | Strategic (AI models) |
| NVIDIA | Strategic (GPU computing) |
| Jeff Bezos (Bezos Expeditions) | Individual investor |
| Intel Capital | Venture capital |
| Parkway Venture Capital | Lead investor |
| Align Ventures | Venture capital |
| ARK Invest | Investment management |
| Amazon Industrial Innovation Fund | Corporate venture |
Alongside the Series B, Figure announced a collaboration agreement with OpenAI to develop next-generation AI models for its humanoid robots.[4] The collaboration also included an agreement to use Microsoft Azure for AI infrastructure, training, and storage.
The collaboration with OpenAI, announced in February 2024, focused on integrating large language models with Figure's robots to enable conversational capabilities and language-grounded reasoning. On March 13, 2024, Figure released a video showing Figure 01 holding a full conversation with a human while performing tasks on a kitchen counter. The system worked by feeding images from the robot's cameras and transcribed speech from onboard microphones to a large multimodal model trained by OpenAI. In the demonstration, a human asked the robot, "Can I have something to eat?" The robot surveyed the objects in front of it, identified an apple, handed it to the person, and then verbally explained its reasoning for the action. The video went viral, accumulating millions of views and drawing widespread attention to the potential for conversational humanoid robots. Unlike scripted demonstrations from other robotics companies, Figure emphasized that the interactions were not pre-programmed; the robot was making real-time decisions based on visual input, verbal commands, and contextual understanding.
In August 2024, Figure 02 launched with OpenAI-powered speech-to-speech conversation as its primary human interaction interface. The robot could engage in natural spoken dialogue while performing tasks.
However, in February 2025, Figure ended its partnership with OpenAI and transitioned to fully in-house AI development.[5] CEO Brett Adcock explained the decision in several ways. He wrote that "LLMs are getting smarter yet more commoditized" and that "for us, LLMs have quickly become the smallest piece of the puzzle." He told TechCrunch: "We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI. We can't outsource AI for the same reason we can't outsource our hardware."[5] The company announced it had achieved a "major breakthrough" in fully end-to-end robot AI built entirely in-house, which became the Helix model.
Around the same time, OpenAI was rebuilding its own robotics division and had filed a trademark application mentioning "humanoid robots" among other technologies.
On March 15, 2025, Figure introduced BotQ, a high-volume manufacturing facility designed as a state-of-the-art production center capable of producing up to 12,000 humanoid robots annually in its first generation line, with plans to scale to 100,000 robots over four years.[7]
In April 2025, Fortune raised questions about the scope of Figure's BMW pilot versus public claims. BMW described limited, off-hours trials during that period. Figure disputed the article and indicated legal action.
In May 2025, reports emerged that UPS was in discussions with Figure AI about deploying humanoid robots in its logistics operations. The talks had begun as early as 2023 and progressed through 2024, focusing on the potential for Figure 02 robots to perform physically repetitive tasks such as lifting, sorting, and transporting packages in warehouse environments. As of mid-2025, the initiative remained in an exploratory phase.
In September 2025, Figure announced a strategic partnership with Brookfield Asset Management to develop what they described as the "world's largest humanoid pretraining dataset" using real-world environments.[34] Brookfield, a global alternative asset manager with more than $1 trillion in assets under management, owns over 100,000 residential units, more than 500 million square feet of commercial office space, and 160 million square feet of logistics space. The partnership, internally designated Project Go-Big, involves capturing first-person (egocentric) video of people performing everyday tasks across a variety of Brookfield-managed properties. This data is used to train Helix so that humanoid robots can learn how to move, perceive, and act in human-centric spaces without requiring robot-specific demonstration data.
In October 2025, Figure closed its Series C funding round, raising over $1 billion at a $39.5 billion post-money valuation.[10][40] Parkway Venture Capital led the round, with participation from Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Microsoft, the OpenAI Startup Fund, and Jeff Bezos through Bezos Expeditions. Announcing the round, Adcock said: "This milestone is critical to unlocking the next stage of growth for humanoid robots, scaling out our AI platform Helix and BotQ manufacturing."[40] The valuation represented a roughly 15-fold increase from the $2.6 billion Series B valuation just 20 months earlier. The company reported total funding of approximately $1.85 billion across all rounds. The Series C capital was earmarked for scaling humanoid robot production, expanding the BotQ manufacturing facility, and broadening real-world commercial and consumer deployments.
On October 9, 2025, TIME profiled Figure and reported the reveal of Figure 03, its third-generation humanoid robot redesigned for mass production and home use.[8]
Also in late 2025, CEO Brett Adcock announced that Figure would begin "alpha testing" its humanoid robots in real homes, a timeline advanced by approximately two years due to confidence in Helix's capabilities.[36] The alpha program targeted select participants in the Bay Area for initial in-home deployments.
In November 2025, Figure published detailed results from its deployment at BMW's Spartanburg plant.[9] Over a period of roughly 10 months, a Figure 02 robot operated on the production line during 10-hour shifts, Monday through Friday. The robot performed sheet-metal loading, a pick-and-place task in which sheet-metal parts are removed from racks and placed onto welding fixtures with a tolerance of 5 millimeters within 2 seconds. The deployment achieved the following results:
| Metric | Result |
|---|---|
| Vehicles contributed to | 30,000+ BMW X3s |
| Sheet-metal parts loaded | 90,000+ |
| Operational hours | 1,250+ |
| Steps walked | 1.2+ million (approximately 200+ miles) |
| Cycle time | 84 seconds total, 37-second load time |
| Placement accuracy target | >99% success rate per shift |
| Speed improvement | 400% over initial deployment |
| Success rate improvement | 7x over initial deployment |
| Daily capacity | 1,000+ operations |
The deployment also revealed that the forearm was the primary hardware failure point, leading Figure to completely redesign the wrist electronics for the next-generation Figure 03, eliminating distribution boards and dynamic cabling.
In late January 2026, Figure announced Helix 02, a major upgrade to its AI system that introduced full-body autonomy.[11] Helix 02 added a new foundation layer called System 0 to the existing architecture, enabling the robot to use its entire body as a tool rather than relying solely on hand manipulation.
In February 2026, BMW announced that it would expand humanoid robot deployments to its Plant Leipzig in Germany, marking the first time humanoid robots entered a European automotive production environment.[13] While the Leipzig deployment uses a different robot (AEON by Hexagon Robotics), BMW credited the success of its Figure 02 pilot at Spartanburg as the foundation for its broader "Physical AI" strategy and established a new Center of Competence for Physical AI in Production to coordinate global integration of AI and robotics. The Leipzig deployment focuses on battery assembly and component production, with a full pilot phase planned for summer 2026.
On March 9, 2026, Figure published a "Helix 02 Living Room Tidy" demonstration in which a single Figure 03 robot cleaned a cluttered living room using only Helix 02 control. The robot autonomously sprayed water onto a surface and wiped it with a towel, collected scattered toys and blocks into containers held with both hands, tossed a pillow back onto a couch, and pressed buttons on a TV remote after reorienting it in-hand. The demonstration also included whole-body efficiency strategies such as tucking containers under the arm and side-stepping through narrow gaps between furniture. According to Figure, no new algorithms or specialized programming were required for the task; Helix 02 learned the behavior by adding new data to the same neural policy used for prior tasks.[54]
In late March 2026, reports emerged that Adcock was also working on a new startup called Hark, focused on AI-powered consumer devices. Hark plans to combine foundation models, software systems, and custom hardware to create physical products that anticipate user needs.
On April 29, 2026, Figure published "Ramping Figure 03 Production," disclosing that BotQ had delivered more than 350 Figure 03 robots and that the production rate had increased from one robot per day to one robot per hour, a roughly 24-fold throughput improvement in under 120 days.[55] The company reported a battery-line first-pass yield of 99.3% and an end-of-line first-pass yield above 80% (improving weekly), with more than 9,000 actuators produced across over 10 distinct SKUs. The facility's custom manufacturing execution software ran across more than 150 networked workstations, with each robot subject to more than 50 in-process inspection points and over 80 functional verification tests, including multi-limb stress testing and burn-in exercises.[55]
On May 8, 2026, Figure released "Helix-02 Bedroom Tidy," the first publicly demonstrated case of two humanoid robots performing collaborative loco-manipulation under a single learned vision-language-action model. Two Helix 02-equipped robots reset a bedroom in under two minutes, opening doors with whole-body coordination, draping clothing on narrow fixtures, closing a book with bimanual dexterous control, operating a trash can foot pedal in single-leg stance, and coordinating with each other to make a bed using bimanual whole-body motions. According to Figure, each robot reads its partner's intent purely from visual cues, with no centralized planner or explicit message passing between the units.[56]
On May 13, 2026, Figure livestreamed a fleet demonstration in which a team of Figure humanoids ran a continuous package-sorting shift on a conveyor belt. The robots scanned barcodes, picked up boxes, reoriented them, and routed them into the workflow at roughly three to four seconds per package, a rate Figure described as matching human performance. The robots used onboard battery management and signaled one another to switch in and out of the line for autonomous charging when batteries ran low. Adcock posted on X: "Watch a team of humanoid robots running a full 8-hr shift at human performance levels. This is fully autonomous running Helix-02."[57] The demonstration extended well beyond its initial 8-hour scope, with the same fleet running for more than 50 hours nonstop in the days that followed. In a Bloomberg Television interview on May 15, 2026, Adcock said: "There's absolutely no teleoperation into this."[58]
By May 17, 2026, an individual Figure robot named "Jim" had completed 81 continuous hours of autonomous package sorting at a Figure logistics facility, processing 101,391 packages without human intervention or teleoperation.[59] During the first 24 hours of the run, Figure reported no errors or interruptions. Each package was processed in roughly three to four seconds, with the robot visually recognizing boxes, identifying label orientation, and routing items using only on-robot compute and the Helix 02 model.[59]
As of mid-2026, Figure AI employs approximately 400 to 500 people and continues to scale its operations. The company is focused on three core areas: expanding humanoid robot deployments into homes and commercial operations, ramping production at BotQ, and building next-generation GPU infrastructure to accelerate training and simulation for the Helix platform.
The Figure 01, announced in March 2023, was the company's first-generation humanoid robot. It was designed as a general-purpose bipedal robot for manual labor tasks, initially targeting the logistics and warehousing sectors.
| Specification | Figure 01 |
|---|---|
| Height | 168 cm (5 ft 6 in) |
| Weight | 60 kg (132 lb) |
| Payload capacity | 20 kg (44 lb) |
| Degrees of freedom | 40+ |
| Walking speed | 1.2 m/s (4.3 km/h) |
| Battery life | Up to 5 hours |
| Actuators | Custom electromechanical |
| Hands | Four fingers and opposable thumb per hand |
Key milestones for Figure 01 included:
The Figure 01 served as the company's proof-of-concept platform. Lessons from its development informed the design of the Figure 02.
The Figure 02 was unveiled in August 2024 after roughly 18 months of development. It represented a substantial hardware and software upgrade over the Figure 01, with improvements to computing power, battery capacity, dexterity, and sensor systems. It was the first Figure robot designed for sustained commercial deployment and featured deeply integrated AI systems.
| Specification | Figure 02 |
|---|---|
| Height | 168-170 cm (5 ft 6 in-5 ft 7 in) |
| Weight | 70 kg (154 lb) |
| Payload capacity | 20 kg (44 lb) |
| Hand payload | 25 kg (55 lb) per hand |
| Degrees of freedom | 41 |
| Walking speed | 1.2 m/s |
| Battery capacity | 2.25 kWh lithium-ion (torso-integrated) |
| Battery life | Up to 5-10 hours active use |
| Charge time | ~1.5 hours (autonomous docking) |
| Cameras | 6 RGB cameras |
| Hand DoF | 16 per hand |
| Compute | NVIDIA RTX GPU modules (3x Figure 01) |
| Audio | Built-in speakers and microphones |
| Power | Fully electric with integrated cabling |
The Figure 02's fourth-generation hands feature 16 degrees of freedom each, with five fingers capable of manipulating tools and carrying loads up to 25 kg. The battery capacity increased by 50% over the Figure 01, and the power source was relocated to the torso to improve balance and center of gravity. AI inference capability tripled compared to the previous generation, powered by NVIDIA RTX GPU-based onboard modules.
The Figure 02 became the company's primary commercial deployment platform. It was deployed at BMW's Spartanburg plant and in logistics warehouse operations (described in detail below). The robot demonstrated the ability to learn new tasks through the Helix VLA system, performing sheet-metal loading operations with increasing accuracy and speed over the course of deployment.
The Figure 03 was announced October 9, 2025 as Figure's third-generation humanoid, designed specifically for home and consumer environments.[8] It marked a strategic shift from purely industrial applications toward domestic use and represented a complete hardware and software redesign optimized for mass production and safe home use. Figure described it as "designed for Helix, the home, and the world at scale."[8]
| Specification | Figure 03 |
|---|---|
| Height | 168 cm (5 ft 6 in) |
| Weight | 61 kg (134 lb), 9% lighter than Figure 02 |
| Payload capacity | 20 kg (44 lb) |
| Battery life | Up to 5 hours |
| Charging | Wireless inductive at 2 kW (via foot charging coils) |
| Data offload | 10 Gbps mmWave |
| Actuators | 2x faster speeds vs. Figure 02 |
| Tactile sensors | Custom, detecting forces as small as 3 grams |
| Camera frame rate | 2x Figure 02 |
| Camera latency | 1/4 of Figure 02 |
| Camera field of view | 60% wider per camera vs. Figure 02 |
| Palm cameras | Embedded in each hand |
| Audio | Speaker 2x larger, 4x more powerful than Figure 02 |
| Exterior | Soft textile covering, multi-density foam padding |
| Battery safety | UN38.3 certified, multi-layer protection |
| Target price | Approximately $20,000 (unconfirmed) |
The Figure 03 introduced several features tailored for household environments. Its exterior uses soft textile coverings instead of hard machined surfaces, with multi-density foam padding at pinch points for safe operation around people. The soft goods are fully washable and tool-free removable. Embedded palm cameras provide grasping feedback, and proprietary first-generation tactile sensors in the fingertips can detect forces as light as 3 grams, roughly the weight of a paperclip.[8] Other safety features for home use include enhanced audio (2x speaker size, 4x power), improved microphone positioning for natural conversation, and a UN38.3-certified battery with multiple redundant safety layers.
Wireless inductive charging through coils in the robot's feet enables near-continuous operation: the robot can top off its battery by simply standing on a charging pad, eliminating the need for manual plug-in. A 10 Gbps mmWave data link allows rapid fleet data uploads for continuous machine learning improvement.
Demonstrated household tasks include folding towels, loading dishwashers, clearing tables, loading laundry, sorting laundry, and navigating rooms while responding to voice commands. Broader home availability is targeted for late 2026, with select partner testing underway.
The Figure 03 is also the model optimized for high-volume manufacturing at the BotQ facility, with design changes (such as replacing CNC-machined parts with die-cast and injection-molded components) aimed at reducing production cost and time. Figure 03 is also slated to replace Figure 02 in future industrial rollouts at facilities like BMW.
| Model | Release Date | Height | Weight | Payload | Runtime | Key Features |
|---|---|---|---|---|---|---|
| Figure 01 | 2022-2023 | 1.68 m (5'6") | 60 kg (132 lb) | 20 kg (44 lb) | 5 hours | First bipedal prototype, five-finger hands |
| Figure 02 | August 2024 | 1.70 m (5'7") | 70 kg (155 lb) | 25 kg per hand | 5-10 hours | 6 cameras, conversational AI, 16 DoF hands, 2.25 kWh battery |
| Figure 03 | October 2025 | 1.68 m (5'6") | 61 kg (134 lb) | 20 kg | 5 hours | Mass production ready, home-safe design, wireless charging, tactile sensors |
Helix is Figure's proprietary vision-language-action model (VLA), unveiled in February 2025.[6] It is the AI system that powers all of Figure's current robots, replacing the earlier OpenAI-based models. Helix was developed entirely in-house by Figure's AI team.
Helix uses a dual-system architecture inspired by cognitive science concepts of fast and slow thinking:
System 2 is a 7-billion-parameter vision-language model that operates at 7 to 9 Hz. It handles high-level scene understanding, language comprehension, and task planning. System 2 is built on an open-source, open-weight VLM pretrained on internet-scale data, then specialized for robotic applications. It processes camera images and robot state information to produce a latent semantic vector that encodes the current task context.
System 1 is an 80-million-parameter transformer-based visuomotor policy that operates at 200 Hz. It translates the latent representations from System 2 into continuous motor commands for the robot's joints. System 1 uses a fully convolutional, multi-scale vision backbone initialized from simulation-based pretraining. It outputs control signals for 35 degrees of freedom simultaneously, including individual finger movements, wrist poses, head gaze, and torso orientation.
The two systems communicate through a continuous latent vector. System 2 provides semantic objectives at its slower rate, while System 1 maintains fast closed-loop control at 200 Hz to handle the real-time dynamics of physical manipulation.
Helix is trained on approximately 500 hours of high-quality teleoperation data collected across multiple robots and operators.[6] A VLM generates hindsight text instructions by analyzing video segments and asking: "What instruction would have produced this observed action?" This auto-labeling process eliminates the need for humans to manually annotate every demonstration.
The training uses end-to-end regression from raw pixels and text to continuous joint actions. No task-specific fine-tuning is required; a single set of model weights handles all tasks.
Helix introduced several firsts for VLA models:
Helix runs entirely on the robot's embedded low-power GPUs, with no external compute required during operation.
In late January 2026, Figure released Helix 02, extending neural network control from the upper body to the entire robot and enabling what Figure calls "full-body autonomy."[11] Where the original Helix controlled primarily the upper body, Helix 02 controls the entire robot from a single unified neural network. Helix 02 added a third layer to the architecture:
| System | Function | Frequency | Description |
|---|---|---|---|
| System 2 (S2) | Semantic reasoning | 7-9 Hz (deliberative) | Interprets scenes, understands language, sequences multi-step behaviors |
| System 1 (S1) | Visuomotor policy | 200 Hz | Translates perception into full-body joint targets using a transformer-based model |
| System 0 (S0) | Whole-body controller | 1,000 Hz (1 kHz) | Handles balance, contact, and coordination across the entire body |
System 0 is the key innovation in Helix 02. It is a 10-million-parameter neural network operating at 1 kHz, trained entirely in simulation across more than 200,000 parallel environments with domain randomization for sim-to-real transfer. The training data includes over 1,000 hours of joint-level retargeted human motion data. According to Figure, System 0 replaces 109,504 lines of hand-engineered C++ code with a single neural prior for stable, natural motion.
System 1 in Helix 02 accepts inputs from head cameras, palm cameras, fingertip tactile sensors, and full-body proprioception. It is the first demonstrated neural policy to utilize palm cameras and tactile sensing modalities simultaneously, outputting complete joint-level control across legs, torso, head, arms, wrists, and individual fingers.
System 2 has been expanded to enable multi-step behavioral commands. Where the original Helix could process single-object tasks ("pick up the cup"), Helix 02 can handle complex chained instructions such as "Walk to the dishwasher and open it."
With all three layers working together, Helix 02 enables integrated locomotion and manipulation. In its keynote demonstration, Helix 02 autonomously unloaded and reloaded a dishwasher across a full-sized kitchen in a continuous four-minute sequence involving 61 sequential loco-manipulation actions. The task required motor control spanning four orders of magnitude, from millimeter-precision finger movements to room-scale locomotion, with no human intervention or resets. When the robot's hands were occupied, it used other body parts to interact with the environment, closing a drawer with its hip and lifting the dishwasher door with its foot. Figure described this as the longest-horizon, most complex task completed autonomously by a humanoid robot at the time of announcement.
Additional demonstrated capabilities include:
Helix 02 runs entirely on embedded GPUs within the Figure 02 and Figure 03 robots, with low power consumption, enabling real-world deployment without reliance on external computing resources or cloud connectivity.
Following the January 2026 launch, Figure progressively released longer-horizon demonstrations of Helix 02 in unstructured home and warehouse environments using the same neural policy. On March 9, 2026, the "Helix 02 Living Room Tidy" video showed a single Figure 03 robot autonomously cleaning a cluttered living room, including spraying and wiping a surface, sorting toys and blocks into containers held with both hands, tossing a pillow onto a couch, and pressing buttons on a TV remote after reorienting it in-hand.[54] On May 8, 2026, "Helix-02 Bedroom Tidy" showed two robots cooperatively resetting a bedroom in under two minutes - opening doors, hanging clothes, closing a book with bimanual control, operating a foot pedal trash can on one leg, and jointly making a bed. The bedroom demo is, according to Figure, the first published case of two humanoids performing collaborative loco-manipulation under a single learned vision-language-action policy, with no centralized planning or explicit inter-robot messaging.[56] In May 2026, Helix 02 also served as the policy backbone for Figure's 8-hour, 50-hour, and 81-hour autonomous package-sorting runs (see Logistics and warehousing).
Figure's first and most prominent commercial deployment has been with BMW. In January 2024, Figure signed a commercial agreement with BMW Manufacturing to deploy humanoid robots at BMW Group Plant Spartanburg in South Carolina. The collaboration represents one of the first sustained commercial deployments of humanoid robots in automotive manufacturing.
The deployment followed a staged approach:
Tasks performed at Spartanburg included:
The deployment focused on three critical performance measurements: cycle time (84-second total cycle with 37-second load time), placement accuracy (greater than 99% success rate per shift), and intervention rate (targeting zero human pauses or resets per shift).
Following the successful Spartanburg pilot, BMW announced in early 2026 that it would expand humanoid robot deployment to Plant Leipzig in Germany (using Hexagon Robotics' AEON robot), marking the first European pilot for humanoid robots in the BMW production network.[13] The Leipzig deployment focuses on battery assembly and component production, with a full pilot phase planned for summer 2026. BMW also established a "Center of Competence for Physical AI in Production" to accelerate integration of AI and robotics across its global manufacturing operations, with plans to evaluate Figure 03 for additional use cases in future deployments.
In February 2025, Figure demonstrated a second commercial application: logistics package handling and sorting. CEO Brett Adcock stated that it took only 30 days to deploy Figure 02 robots at an unnamed warehouse for the company's second customer.
In unedited hour-long demonstrations, Figure 02 robots sorted packages at warehouse facilities, processing each package in approximately 4.05 seconds (a 20% improvement over earlier performance) with a 95% barcode scanning success rate. The robots handled diverse package types, from rigid boxes to deformable bags, reorienting items to ensure correct label positioning for downstream scanning. Over three months of operation, the system's capabilities improved to approach human-level dexterity and speed for package sorting tasks.
In May 2026, Figure used logistics package sorting as the testbed for its longest publicly demonstrated autonomous humanoid runs to date, all running on the Helix 02 model. On May 13, 2026, a multi-robot fleet completed a continuous eight-hour shift on a conveyor belt, processing barcoded packages at roughly three to four seconds per package while autonomously rotating units off the line for self-charging when batteries ran low.[57] The livestream was subsequently extended for over 50 hours of continuous operation, which Adcock told Bloomberg Television on May 15, 2026 had occurred with no teleoperation.[58] On May 17, 2026, a single Figure robot designated "Jim" completed an 81-hour autonomous run sorting 101,391 packages at a Figure logistics facility, again with no human interventions and using only on-robot compute and the Helix 02 model.[59] Figure characterized these runs as evidence that humanoid robots can sustain industrial-shift workloads at human throughput without offline planning or remote operators.
Reports in 2025 indicated that UPS was exploring the potential deployment of Figure humanoid robots in its logistics operations. Discussions between the two companies reportedly began in 2023, with the focus on using robots for physically repetitive warehouse tasks such as lifting, sorting, and transporting packages. As of mid-2025, the initiative remained in an exploratory phase, with no confirmed deployment dates.
As part of the February 2024 funding round, Figure established a partnership with Microsoft for AI infrastructure. Figure uses Microsoft Azure for training its AI models, running simulations, and storing data generated by robot deployments.
In September 2025, Figure announced a strategic partnership with Brookfield Asset Management to build what they described as the "world's largest humanoid pretraining dataset."[34] Through the initiative, internally designated Project Go-Big, Figure captures first-person video of people performing everyday tasks across Brookfield's real estate portfolio. Brookfield's assets include over 100,000 residential units, more than 500 million square feet of commercial office space, and 160 million square feet of logistics space. The data collected feeds directly into training the Helix model, allowing robots to learn navigation and manipulation behaviors from human video without requiring robot-specific demonstrations.
In March 2025, Figure announced BotQ, a dedicated high-volume manufacturing facility for humanoid robots.[7] The company described it as "the highest volume humanoid production line in the world" at the time of announcement. Located in California, BotQ represents Figure's strategy of vertically integrating the production of humanoid robots rather than relying on contract manufacturers.
| Metric | Capacity |
|---|---|
| Initial annual production | 12,000 robots |
| 4-year target | 100,000 robots |
| Actuator production capacity | 3,000,000 units over 4 years |
| Manufacturing processes | Injection molding, die-casting, metal injection molding, stamping |
| Part production time reduction | From over 1 week (CNC) to under 20 seconds (tooled processes) |
The facility represented a fundamental shift in Figure's manufacturing approach. During prototype development, the company relied heavily on CNC machining, which was slow and expensive. For BotQ, Figure transitioned to tooled processes: injection molding, die-casting, metal injection molding, and stamping. Parts that previously required over a week of CNC machining can now be manufactured in under 20 seconds using steel molds. The design team also consolidated parts to reduce total component count and assembly complexity.
The facility features:
BotQ uses a hybrid workforce of humans and Figure humanoid robots. The robots operate on the assembly lines as material handlers and assist with component assembly, using the Helix AI system. As production ramps, Figure plans to expand the proportion of robot-assisted assembly.
The Figure 03 is the primary model designed for BotQ production, with its hardware optimized for affordability and high-volume manufacturing.
On April 29, 2026, Figure published detailed metrics on the BotQ ramp through the first 120 days of Figure 03 production:[55]
| Metric | Status (April 29, 2026) |
|---|---|
| Figure 03 robots delivered | 350+ |
| Throughput improvement (120 days) | ~24x |
| Production rate | 1 robot per hour |
| End-of-line first-pass yield | >80% (improving weekly) |
| Battery line first-pass yield | 99.3% |
| Actuators produced | 9,000+ |
| Actuator SKUs | 10+ distinct |
| In-process inspection points | 50+ per robot |
| Functional verification tests | 80+ per robot |
| Networked workstations | 150+ |
Figure described the ramp as demonstrating the per-hour cycle time required to hit its 12,000-units-per-year first-generation production target, and characterized the end-of-line tests, which include multi-limb stress testing and burn-in exercises, as enabling rapid identification of regressions during high-volume manufacturing.
Figure pursues deep vertical integration across its technology stack. The company designs and builds its own motors, actuators, sensors, battery packs, structural components, firmware, control software, and AI models in-house. Adcock has compared this approach to how Tesla vertically integrates vehicle manufacturing and how Archer Aviation integrated aircraft systems.
The company's rationale for the humanoid form factor is practical: human environments (homes, factories, warehouses) are already designed for human-shaped bodies. A robot that matches human dimensions can use existing tools, navigate existing spaces, and operate existing equipment without requiring environmental modifications.
Figure operates on six-month hardware and software release cycles, emphasizing rapid iteration. The company uses NVIDIA GPU clusters (including ND H100 instances provided through its Microsoft partnership) for training the Helix models, and NVIDIA's simulation and inference tools for development and testing.
Key technological components include:
Safety features across Figure's robots include redundant torque sensors, active fall mitigation systems, and certified override mechanisms. Rob Gruendel, the company's Robotics Safety Lead, participates in ANSI/ISO standards committees for robot safety.
| Round | Date | Amount | Valuation | Lead Investors |
|---|---|---|---|---|
| Seed | 2022 | ~$100M | N/A | Bold Capital Partners (Brett Adcock self-funded) |
| Series A | May 2023 | $70M | $500M | Parkway Venture Capital |
| Series A Extension | July 2023 | $9M | $350M | Intel Capital |
| Series B | February 2024 | $675M | $2.6B | Microsoft, OpenAI, NVIDIA, Jeff Bezos |
| Series C | October 2025 | >$1B | $39.5B | Parkway VC, Brookfield, NVIDIA |
Total funding exceeds $1.85 billion as of October 2025.
The growth trajectory of Figure AI's valuation illustrates the intense investor interest in humanoid robotics:
| Date | Valuation | Multiple from Previous |
|---|---|---|
| May 2023 (Series A) | $500M | Baseline |
| July 2023 (Extension) | $350M | 0.7x (down round on extension) |
| February 2024 (Series B) | $2.6B | ~7.4x from Series A Extension |
| October 2025 (Series C) | $39.5B | ~15.2x from Series B |
Notable investors across all rounds include:
Figure operates in an increasingly competitive humanoid robotics market, competing with established and emerging companies across industrial and consumer segments.
| Company | Robot | Market Focus | Notable Features |
|---|---|---|---|
| Tesla | Optimus (Gen 3) | Manufacturing, consumer | Targeting mass production at scale; plans for 1 million units/year by 2030; price target $20,000-$30,000; 11 DoF per hand; leverages FSD neural networks |
| Boston Dynamics | Atlas (Electric) | Industrial | Decades of robotics research; 56 DoF, 50 kg lift capacity, production version launched at CES 2026; confirmed 2026 deployments with Hyundai and Google DeepMind; estimated $140,000+ per unit |
| Agility Robotics | Digit | Warehousing | $178 million raised; focused on logistics; $550 million valuation; first US humanoid mass production facility (10,000/year capacity) |
| 1X Technologies | NEO Gamma | Home use | $140 million raised; backed by OpenAI; targeting home and commercial use |
| Apptronik | Apollo | Manufacturing | $431 million raised; partnered with NASA and Mercedes-Benz |
| Unitree Robotics | G1, H1 | Multiple sectors | Chinese manufacturer; lower price points; rapidly iterating hardware |
| UBTech Robotics | Walker S | Automotive | Publicly traded; $36.5 billion market cap; deployed in automotive factories in China |
| Hexagon Robotics | AEON | Automotive | Selected by BMW for Leipzig plant deployment (2026) |
Figure differentiates itself through the combination of its proprietary Helix AI system, its early commercial deployments (BMW and logistics customers), its high funding levels, and its BotQ manufacturing infrastructure. CEO Brett Adcock has publicly positioned Figure as having "the most advanced humanoid" in the industry, a claim directly competing with Tesla's Optimus program and Chinese robotics companies like Unitree and UBTech.
The humanoid robotics industry saw significant acceleration in 2025 and early 2026. Several competitive dynamics are shaping the market.
Tesla Optimus remains Figure's most prominent competitor in terms of public attention. At Tesla's October 2025 event, Elon Musk demonstrated Optimus Gen 3 performing complex tasks such as cooking and household cleaning, learned autonomously through observation. Tesla targets a price point of $20,000 to $30,000 per unit. However, during Tesla's Q4 2025 earnings call in January 2026, Musk acknowledged that despite prior claims of 1,000+ deployed units, no Optimus robots were performing "useful work" in factories at that time.
Boston Dynamics Atlas transitioned to a fully electric design in 2024 and launched its production version at CES 2026 with 56 degrees of freedom, 50 kg lift capacity, and confirmed 2026 deployments with Hyundai and Google DeepMind. Atlas is positioned as an enterprise-grade industrial robot with an estimated price point of $140,000 or more per unit.
Funding momentum across the sector has been substantial. In a single week in early 2026, robotics companies Mind Robotics, Rhoda AI, Sunday, and Oxa collectively raised over $1.2 billion. Combined with Figure's Series C and SkildAI's $1.4 billion round, 2026 is on pace for over $20 billion in total robotics funding.
Analysts project substantial growth for the humanoid robotics market over the coming decade:
| Source | Projection | Timeframe |
|---|---|---|
| Goldman Sachs | $38 billion total addressable market | By 2035 |
| Goldman Sachs | 250,000+ humanoid shipments (base case) | By 2030 |
| Goldman Sachs | 1.4 million humanoid shipments | By 2035 |
| ARK Invest | $10-$24 trillion total addressable market | Long-term (wide adoption) |
| Morgan Stanley | $5 trillion market | By 2050 |
Goldman Sachs revised its humanoid robot market projections upward by more than sixfold from an earlier estimate of $6 billion, citing advances in AI, particularly robotic large language models, as the primary driver of the revision.[50]
As of early 2026, at least a dozen companies are actively building, testing, or deploying humanoid robots across factories, warehouses, and homes. Figure AI's CEO Brett Adcock has stated: "In the next 10 years, maybe under 10 years, the biggest company in the world will be a humanoid robot company. Every home will have a humanoid."
Figure generated its first revenue in December 2024 through its BMW deployment. The company's commercial approach includes both Hardware as a Service (HaaS) and Robotics as a Service (RaaS) models, with multi-year enterprise contracts. For the consumer market, Figure plans direct sales combined with software subscriptions.
The company has stated a target unit price of under $20,000 for the Figure 03, which would make it competitive with Tesla's projected Optimus pricing. At scale, Figure is targeting two major customer segments: commercial and industrial operations (manufacturing, logistics, retail) in the near term, and consumer households by approximately 2030. The company has cited a global addressable market of 2.3 billion households, with more than 700 million elderly individuals potentially requiring care assistance.
Brett Adcock (born April 6, 1986) is an American technology entrepreneur and billionaire. He grew up on a corn and soybean farm near Moweaqua, Illinois. He began building web companies at age 16, including an e-commerce site focused on outdoor electronics and a content site called Street of Walls (a finance career website). He graduated as valedictorian of Central A&M High School and earned a Bachelor of Science in Business Administration from the University of Florida in 2008 (attended 2004 to 2008).
Prior to Figure AI, Adcock founded:
Adcock's stated vision for Figure is to make labor optional by providing every human with a personal humanoid robot, starting with industrial applications and expanding to home use. As of June 2024, Forbes estimated Adcock's net worth at $1.4 billion.
| Name | Title | Background |
|---|---|---|
| Brett Adcock | Founder and CEO | Founded Vettery, Archer Aviation, Cover, Hark |
| Jerry Pratt | Former CTO | 20 years at IHMC; DARPA Robotics Challenge; left Figure in 2025 to co-found Persona AI |
| Dana Votypka | Chief Financial Officer | Corporate finance and operations |
| Bob Klunk | Chief Operating Officer | Manufacturing and operations |
| Lee Randaccio | VP of Growth | Business development and partnerships |
Note: Jerry Pratt departed Figure in 2025 and co-founded Persona AI with Nic Radford, another veteran roboticist.
Media coverage has highlighted both the rapid progress and the challenges facing general-purpose humanoids. The March 2024 OpenAI demonstration video brought Figure widespread public attention, with millions of views and extensive coverage across technology and mainstream outlets. The demonstration was widely regarded as one of the most convincing real-time humanoid interactions shown publicly at that time.
In April 2025, Fortune raised questions about the scope of Figure's BMW pilot versus public claims; BMW described limited, off-hours trials during that period, while Figure disputed the article and indicated legal action.[32] A subsequent TIME feature in October 2025 profiled the company's long-term consumer ambitions and the debate over data-driven training, dexterity, safety, and societal impacts. The company's November 2025 publication of detailed BMW deployment metrics (30,000+ vehicles, 90,000+ parts, 1,250+ hours) provided concrete operational data that partially addressed earlier skepticism about the deployment's scope.[9]
The broader discourse around Figure AI reflects a tension common in humanoid robotics: the gap between impressive demonstrations and sustained, reliable commercial deployment. Critics note that even the most advanced humanoid robots remain far from the level of general-purpose capability needed for autonomous home use, while supporters point to the rapid pace of improvement and the growing volume of real-world deployment data.
Figure AI is an American artificial intelligence and robotics company that builds general-purpose humanoid robots and the AI that controls them. Founded in May 2022 by Brett Adcock in Sunnyvale, California, the company designs its hardware and its in-house vision-language-action model, Helix, so that a single robot platform can perform manufacturing, logistics, and household tasks. Figure has built three robot generations (Figure 01, Figure 02, and Figure 03) and is one of the most valuable private robotics companies in the world.[1][40]
Figure AI reached a $39 billion post-money valuation when it announced a Series C round of more than $1 billion in committed capital in September 2025.[10][40] The round was led by Parkway Venture Capital, with participation from Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Salesforce, T-Mobile Ventures, and Qualcomm Ventures.[40] The valuation marked a roughly 15-fold increase from the $2.6 billion at which the company raised its $675 million Series B in February 2024, and Figure reports approximately $1.85 billion in total funding across all rounds.[4][40]
Figure AI was founded by Brett Adcock, a serial entrepreneur who previously co-founded the recruiting marketplace Vettery (acquired by the Adecco Group in 2018) and the eVTOL aircraft company Archer Aviation.[2] Adcock self-funded Figure's early operations with approximately $100 million of his own capital and recruited engineers from companies including Boston Dynamics, Tesla, Apple, and Google DeepMind.[14]
Helix is Figure's proprietary vision-language-action model, announced in February 2025 and developed entirely in-house after Figure ended its partnership with OpenAI.[5][6] It uses a dual-system design: a slower vision-language model (System 2) for scene understanding and language, and a fast 200 Hz visuomotor policy (System 1) for motor control.[6] The 2026 upgrade, Helix 02, added a 1 kHz whole-body controller (System 0) for full-body autonomy, letting a single neural network coordinate balance, locomotion, and dexterous manipulation.[11]
Figure's most prominent commercial deployment was at BMW Group Plant Spartanburg in South Carolina, where a Figure 02 robot loaded sheet-metal parts over roughly 10 months and contributed to the production of more than 30,000 BMW X3 vehicles, moving over 90,000 parts across more than 1,250 operational hours.[9] Figure has also run its robots in logistics package-sorting operations, including a single robot that sorted 101,391 packages over an 81-hour autonomous run in May 2026.[59] BMW separately expanded humanoid trials to Plant Leipzig in Germany in 2026.[13]
No. Figure announced a collaboration with OpenAI alongside its February 2024 Series B, but ended the partnership in February 2025 to develop AI fully in-house.[4][5] Brett Adcock told TechCrunch the company decided that "to solve embodied AI at scale in the real world, you have to vertically integrate robot AI," and that the company could not "outsource AI for the same reason we can't outsource our hardware."[5] The OpenAI Startup Fund nonetheless continued to participate as a financial investor in later rounds.
| Date | Event |
|---|---|
| May 2022 | Figure AI founded by Brett Adcock; self-funded with ~$100 million |
| March 2023 | Figure 01 publicly announced; company emerges from stealth |
| May 2023 | $70 million Series A led by Parkway Venture Capital ($500M valuation) |
| July 2023 | $9M Series A extension from Intel Capital ($350M valuation) |
| October 2023 | First dynamic walking demonstrations |
| January 2024 | Commercial agreement signed with BMW Manufacturing |
| February 2024 | $675 million Series B ($2.6B valuation); OpenAI collaboration announced |
| March 2024 | Figure 01 conversational AI demonstration with OpenAI integration |
| August 2024 | Figure 02 unveiled |
| November 2024 | 400% speed improvement demonstrated at BMW plant |
| December 2024 | First revenue generated |
| February 2025 | OpenAI partnership ended; Helix VLA model announced; logistics deployment begins |
| March 2025 | BotQ manufacturing facility announced (12,000 units/year capacity) |
| April 2025 | Fortune raises questions about BMW pilot scope |
| May 2025 | UPS discussions surface |
| September 2025 | Brookfield partnership and Project Go-Big announced |
| October 2025 | >$1 billion Series C ($39.5B valuation); Figure 03 announced for home and consumer use |
| November 2025 | Detailed BMW deployment metrics published |
| January 27, 2026 | Helix 02 released with full-body neural control (System 0/1/2) |
| February 2026 | BMW expands humanoid robot deployment to Plant Leipzig, Germany (AEON robot) |
| March 9, 2026 | Helix 02 Living Room Tidy demo released |
| March 2026 | Reports of Adcock's Hark startup emerge |
| April 29, 2026 | BotQ ramp milestone: 350+ Figure 03 delivered, 1 robot/hour rate, ~24x throughput in 120 days |
| May 8, 2026 | Helix-02 Bedroom Tidy demo - first dual-humanoid collaborative loco-manipulation under a single learned policy |
| May 13, 2026 | 8-hour autonomous package-sorting livestream by Figure fleet on Helix 02 |
| May 15, 2026 | 50+ hour nonstop autonomous run confirmed by Adcock on Bloomberg TV |
| May 17, 2026 | Single Figure robot "Jim" completes 81-hour autonomous run sorting 101,391 packages |