Figure 03
Last reviewed
Sources
36 citations
Review status
Source-backed
Revision
v7 ยท 7,283 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
36 citations
Review status
Source-backed
Revision
v7 ยท 7,283 words
Add missing citations, update stale details, or suggest a clearer explanation.
Figure 03 is a third-generation general-purpose humanoid robot developed by Figure AI and unveiled on October 9, 2025, as the company's first robot designed primarily for the home and engineered from the ground up for high-volume manufacturing. It stands 1.68 meters (5 ft 6 in) tall, weighs 60 kilograms, runs on a 2.3 kWh battery for roughly five hours per charge, and is controlled by Helix, Figure AI's proprietary vision-language-action model. By May 2026 Figure AI had delivered more than 350 Figure 03 units from its BotQ factory and signed what it described as the first paid commercial-scale humanoid contract, an initial fleet of 40 robots for BMW's Spartanburg plant.[3][19]
| Figure 03 | |
|---|---|
 | |
| General information | |
| Manufacturer | Figure AI |
| Country of origin | United States |
| Year unveiled | October 9, 2025 |
| Predecessor | Figure 02 |
| Status | Pre-production / Limited deployment |
| Target price | Below $20,000 (consumer target) |
| Commercial pricing | ~$25 / robot-operating-hour (BMW Spartanburg) |
| AI system | Helix (VLA model), Helix 02, Vulcan balance policy |
| Manufacturing facility | BotQ (3960 N 1st St, San Jose, California; ~98,700 sq ft) |
| Initial production capacity | 12,000 units / year |
| Website | figure.ai |
Figure 03 represents a significant redesign from its predecessor, the Figure 02, shifting the company's focus from purely industrial applications toward consumer and home environments. Figure AI was founded by Brett Adcock. The robot was named one of TIME's Best Inventions of 2025 and appeared on the cover of the magazine's annual Best Inventions issue, accompanied by an exclusive cover story by journalist Billy Perrigo.[1][2]
The robot is built around four design pillars: enabling Figure AI's proprietary vision-language-action model (VLA), ensuring safe operation in homes, supporting high-volume manufacturing, and scaling commercial deployment. It is powered by Helix, the company's in-house VLA, which enables the robot to perceive its environment, understand natural language commands, and execute complex manipulation tasks in real time. In announcing the robot, Figure stated, "There's no path to scaling humanoid robots without AI," and described Figure 03 as "our first robot engineered from the ground-up for high-volume manufacturing."[3] Figure 03 is designed for mass production at BotQ, Figure AI's vertically integrated manufacturing facility in San Jose, California, with an initial capacity of 12,000 units per year and a stated four-year goal of 100,000 robots.[3]
Figure 03 is the first robot in the Figure lineup designed primarily for the home, replacing the hard mechanical shell of the Figure 02 with a soft, washable textile cladding layered over multi-density foam. It introduces wireless inductive charging, palm cameras, fingertip tactile sensors capable of resolving forces as small as three grams, and a torso-integrated battery pack engineered to humanoid robot battery safety standards. CEO Brett Adcock stated that within ten years he expects "every home will have a humanoid," framing Figure 03 as the platform on which that thesis will be tested.[2][3]
By May 2026, Figure 03 had transitioned from a reveal-stage product into an operational commercial platform. The company had delivered more than 350 units from BotQ, ramped production from one robot per day to one per hour, signed what it described as the first paid commercial-scale humanoid contract with BMW Manufacturing's Spartanburg plant for an initial fleet of 40 Figure 03 units, and demonstrated a full eight-hour autonomous logistics shift running its second-generation Helix 02 VLA. In parallel, multi-robot home demonstrations and a new fault-tolerant balance policy called Vulcan extended the platform toward continuously operating, self-maintaining fleets.[19][32][33][34]
Figure AI was founded in 2022 by Brett Adcock, a serial entrepreneur previously known for founding Archer Aviation (an electric vertical takeoff and landing aircraft company) and Vettery (a hiring marketplace). The company set out to build commercially viable general-purpose humanoid robots capable of operating in environments designed for humans. It is headquartered in San Jose, California, and grew from a small founding team to several hundred employees by the time of the Figure 03 unveiling.[4]
The company's first prototype, Figure 01, emerged from stealth in March 2023. The Figure 01 achieved dynamic bipedal walking in under a year of development, which Figure AI described as one of the fastest turnarounds in humanoid robotics history. Standing 1.68 meters tall and weighing 60 kilograms, Figure 01 demonstrated basic warehouse tasks including walking and box manipulation. It used external cabling and was primarily a research platform rather than a production-ready system.[5]
In August 2024, Figure AI unveiled the Figure 02, its second-generation robot designed for industrial deployment. The Figure 02 featured 16 degrees of freedom per hand, five-fingered hands, a 25-kilogram payload capacity, six RGB cameras, integrated internal cabling (replacing the external wiring of Figure 01), and three times the computing power of its predecessor. It was the first Figure robot to run an onboard vision-language-action model and the first to integrate the battery into the torso rather than a backpack.[6]
A pivotal moment in Figure AI's development came through its commercial partnership with BMW Manufacturing, announced in January 2024. Figure 02 robots were deployed at BMW's plant in Spartanburg, South Carolina, where they participated in automotive production operations. Over an 11-month deployment, the Figure 02 fleet accumulated more than 1,250 operational hours, loaded over 90,000 sheet-metal parts, and contributed to the production of more than 30,000 BMW X3 vehicles.[7]
The deployment generated critical data that directly shaped the Figure 03 design. The forearm emerged as the most common hardware failure point during factory operations. For Figure 03, the engineering team completely re-architected the wrist electronics, eliminating both the distribution board and dynamic cabling. Each wrist's motor controller now communicates directly with the main computer, reducing complexity, improving reliability, and simplifying thermal management.[7]
In 2026, BMW expanded its humanoid robotics program to Europe, with a pilot deployment at its Leipzig plant in Germany, establishing a "Center of Competence for Physical AI in Production" to accelerate global integration of AI and robotics across the BMW Group.[8]
Figure AI initially collaborated with OpenAI beginning in 2024, integrating OpenAI's large language models to enable speech-to-speech communication on the Figure 02. OpenAI also participated as an investor in Figure AI's February 2024 funding round.[9]
However, on February 5, 2025, Figure AI ended the partnership. CEO Brett Adcock cited two primary reasons: first, OpenAI announced plans to resurrect its own robotics program, making the two companies potential competitors; second, Figure AI concluded that building effective embodied AI required vertical integration of the robot's AI systems. Adcock stated, "We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI." Following the split, Figure AI developed its Helix model entirely in-house, and the decision shaped both the AI architecture and the hardware design philosophy of Figure 03.[10]
Unlike Figure 01 (a research platform) and Figure 02 (designed for factory deployment with BMW), Figure 03 is positioned as the company's first robot intended for everyday domestic environments. Brett Adcock has described the strategy in interviews as a phased approach: continue training humanoids in logistics and manufacturing settings where data collection and economics are more tractable, then graduate the platform into the home as autonomy and reliability mature. In an interview with TIME, Adcock said that within ten years he expects "every home will have a humanoid," and Figure has stated that select home deployments to partner households began in late 2025.[2][11]
Figure 03 maintains a humanlike form factor at 1.68 meters (5 feet 6 inches) tall, the same height as its predecessors. It weighs 60 kilograms (132 pounds), making it 9% lighter than the Figure 02 (which weighed approximately 70 kilograms). The reduced mass, combined with a more compact overall volume, enables easier navigation through standard doorways, hallways, and household environments. The robot's maximum walking speed is approximately 1.2 meters per second (4.3 km/h), and its rated payload is 20 kilograms.[3]
A defining design change in Figure 03 is its exterior. Rather than the exposed hard plastic and metal shells used on the Figure 01 and Figure 02, the Figure 03 is covered in soft, washable textiles layered over multi-density foam padding. This covering eliminates pinch points and reduces the risk of injury during close-proximity interaction with people. The soft goods are tool-free removable and fully washable, and they can be customized with different garments, including options made from cut-resistant and durable materials. Figure has emphasized that the textile covering is intended both as a safety measure and as an aesthetic choice to make the robot feel less industrial in domestic settings.[3][2]
The outer cladding is sometimes described in Figure's design literature as Figure-shaped soft goods, referring to garment panels specifically tailored to the robot's joints and sensor cutouts. Each panel is patterned around camera apertures, charging coil locations, and joint flex zones so that the textile can be put on and removed without tools, and so that it does not occlude visual or tactile sensing during normal motion. Color and pattern can be configured per deployment, and Figure has previewed cut-resistant variants intended for kitchen and tool-using work, alongside softer cotton-blend variants intended for child-facing home settings.[2][3]
| Category | Specification | Value |
|---|---|---|
| Physical | Height | 1.68 m (5 ft 6 in) |
| Physical | Weight | 60 kg (132 lbs) |
| Physical | Exterior | Soft washable textiles with multi-density foam |
| Physical | Mass reduction vs. Figure 02 | ~9% lighter |
| Mobility | Total degrees of freedom | 44 |
| Mobility | DOF per hand | 16 |
| Mobility | Max walking speed | 1.2 m/s (4.3 km/h) |
| Manipulation | Payload capacity | 20 kg |
| Manipulation | Fingers per hand | 5 |
| Power | Battery designation | F.03 |
| Power | Battery capacity | 2.3 kWh |
| Power | Runtime | ~5 hours at peak performance |
| Power | Charging | 2 kW wireless inductive (via foot coils) |
| Power | Energy density vs. F.01 | +94% |
| Power | Cost vs. F.02 | -78% |
| Sensors | Cameras | 6 main cameras + 2 palm cameras (8 total) |
| Sensors | Tactile | Custom fingertip sensors (3-gram sensitivity) |
| Sensors | Data offload | 10 Gbps mmWave |
| Actuators | Speed improvement | 2x faster than Figure 02 |
| Actuators | Torque density | Improved Nm/kg over Figure 02 |
| Actuators | Motor type | Frameless BLDC motors |
| AI | Model | Helix VLA (System 1 + System 2; later Helix 02 with System 0) |
| AI | Fault tolerance | Vulcan policy (up to 3 lower-body actuator losses) |
| AI | Compute | Dual embedded low-power GPUs |
| Software | Operating system | Ubuntu Linux (LTS) with real-time components |
| Safety | Battery certifications | UN38.3 (achieved); UL2271 (in process) |
| Manufacturing | Production line capacity | Up to 12,000 robots / year (initial) |
| Manufacturing | Stated 4-year goal | 100,000 robots |
| Manufacturing | Throughput (April 2026) | 1 robot / hour (24x vs. launch) |
Figure 03 features a completely redesigned sensory suite purpose-built to enable the Helix AI system. The camera architecture delivers twice the frame rate, one-quarter the latency, and approximately 60% wider field of view per camera compared to the Figure 02's vision system. The robot carries six main cameras for spatial perception and object recognition, plus an embedded camera in each palm for close-range grasping of occluded objects, bringing the total to eight cameras.[3]
The palm cameras are a notable departure from earlier humanoid designs. By embedding wide-angle cameras in each hand, Helix can maintain in-hand visual feedback even when the head-mounted cameras are blocked, such as when reaching into a cabinet, dishwasher rack, or refrigerator drawer. This redundant close-range view is critical for fine manipulation in cluttered or occluded household scenes.[3]
The audio system was also upgraded substantially. The speaker is twice the size and nearly four times more powerful than the Figure 02's speaker, enabling clearer voice communication. Microphone positioning was improved for better speech recognition clarity in noisy home environments.[3]
Each of Figure 03's hands features 16 degrees of freedom and first-generation proprietary tactile sensors integrated into every fingertip. According to Figure, "Each fingertip sensor can detect forces as small as three grams of pressure," which is roughly the weight of a paperclip resting on a finger. This level of sensitivity enables the robot to distinguish between a secure grip and an impending slip before it occurs, allowing fine-grained, dexterous control over fragile, irregular, or moving objects.[3]
The fingertips themselves were redesigned with softer, more adaptive surfaces that increase the contact area during grasps. This combination of tactile feedback and compliant fingertip geometry allows Figure 03 to handle delicate items such as eggs, glassware, and fabrics without damaging them. Figure has stated that the tactile sensors were designed around three principles: extreme durability, long-term reliability, and high-fidelity sensing, reflecting the constraint that consumer hardware must remain functional after thousands of grasps in unstructured environments.[3][12]
Figure 03 is powered by the F.03 battery, a 2.3-kilowatt-hour pack that delivers approximately five hours of runtime at peak performance. The battery represents a 94% increase in energy density compared to the first-generation F.01 battery and a 78% reduction in cost compared to the F.02 battery. Unlike the F.01, which used a bulky external backpack design, the F.03 battery is integrated directly into the robot's torso. The pack's outer enclosure, formed from stamped steel, die-cast aluminum, and structural adhesives, doubles as a load-bearing component of the torso, removing the need for a separate chassis around the battery and reducing weight.[12]
The F.03 battery is the first humanoid robot battery in the process of being certified to both UN38.3 and UL2271 safety standards. It underwent 23 primary tests covering mechanical stress, thermal cycling, and electromagnetic interference. Safety features include a custom Battery Management System (BMS) with sensors, switches, and fuses that prevent overcharge, overdischarge, and overtemperature conditions. Cell-to-cell wirebond interconnect geometry is tuned to act as a fusible element, providing an additional layer of protection. Multiple safeguards aim to prevent thermal runaway from propagating between cells, including thermally insulative potting and flame-containment design. Fault-injection tests in which a single cell is intentionally heated to thermal runaway demonstrated that the pack prevents external flames.[12]
The robot supports wireless inductive charging through coils embedded in its feet. Rather than requiring manual cable connections, Figure 03 can step onto a wireless charging stand and recharge at 2 kilowatts with active forced-convection cooling. The charging system also supports wireless data offload via 10 Gbps mmWave, enabling fleet-wide learning without physical connectivity and removing the need for tethered docking stations in homes.[3][12]
The Figure 03's actuators are capable of twice the speed of those in the Figure 02, with improved torque density (measured in Newton-meters per kilogram). These improvements enable picking and positioning speeds comparable to human workers. Moving from Figure 02 to Figure 03 required the mechanical and electrical engineering teams to redesign nearly every component with manufacturability and cost in mind, aggressively reducing part count, assembly steps, and any components that were not strictly necessary. Figure has reported that the moving joints are smaller and stronger than those in Figure 02, and that the components in aggregate cost approximately 90% less to manufacture.[2][3]
The robot uses frameless brushless DC motors paired with strain-wave (harmonic) and cycloidal gear systems, depending on the joint. Low-level motor control runs at 200 Hz on the dual embedded GPUs that also host the Helix System 1 policy.[13]
Helix is Figure AI's proprietary vision-language-action model that controls Figure 03. Developed entirely in-house following the end of the OpenAI partnership, Helix unifies perception, language understanding, and learned motor control into a single model. It is the first VLA demonstrated to output high-rate continuous control of an entire humanoid upper body, including wrists, torso, head, and individual fingers.[13]
Helix uses a dual-system architecture inspired by cognitive science's "System 1, System 2" framework:
Gradient backpropagation flows between S1 and S2 through latent vectors, enabling end-to-end training from raw pixels and text to continuous robot actions. Both networks run on the embedded low-power GPUs onboard the robot, with S2 operating asynchronously while S1 maintains a real-time 200 Hz control loop.[13]
The original Helix release was trained on approximately 500 hours of high-quality teleoperated demonstrations collected across multiple robots and operators. This dataset is substantially smaller than those used by many competing VLA approaches; Figure has noted it represents less than 5% of the volume used in some published comparable VLA datasets. An auto-labeling vision-language model generated hindsight instructions from video clips, enriching the training data without additional human annotation effort.[13]
The model uses a single unified set of weights for all behaviors, including picking, placing, drawer operation, refrigerator use, and multi-robot handovers, without task-specific fine-tuning. Key capabilities include:
On January 27, 2026, Figure AI released Helix 02, a major upgrade that extends control from the upper body to the entire robot. Where the original Helix governed wrists, torso, head, and fingers, Helix 02 incorporates walking, balance, and locomotion into the same unified neural network, producing what Figure calls full-body autonomy.[14]
Helix 02 introduces a third tier in the architecture:
Figure has stated that S0 replaces 109,504 lines of hand-engineered C++ control code with a single learned neural prior. In a launch demonstration, Figure 03 running Helix 02 completed a four-minute autonomous kitchen task, unloading and reloading a dishwasher across a full-sized kitchen with no resets and no human intervention. The demonstration consisted of 61 coordinated loco-manipulation actions, including bimanual coordination, in-hand object transfers, and whole-body tool use that engaged the hips and feet alongside the hands and arms. Helix 02 operates across motor scales spanning four orders of magnitude, from millimeter-level finger movements to room-scale navigation.[14]
In April 2026, Figure released a perception-conditioned upgrade to System 0 that fused the existing proprioceptive controller with a real-time 3D spatial representation derived from the robot's onboard stereo cameras. Where the launch version of S0 controlled balance and locomotion using joint and IMU signals alone, the upgraded controller takes RGB imagery as an additional conditioning input and converts it into a continuously updated occupancy and surface model. The result is that the same neural prior that handles flat-ground walking can also handle stair climbing and uneven terrain, with the policy trained entirely in simulation and transferred zero-shot to the real Figure 03.[19][33]
In April 2026, Figure publicly demonstrated Vulcan 1.0, a neural balance policy layered on top of Helix 02 that is explicitly trained to tolerate lower-body hardware failures. In Figure's tests, a Figure 03 unit could lose up to three lower-body actuators or joints and continue to maintain balance, walk in a limp mode, and autonomously navigate to a designated repair bay without human intervention. The company has framed Vulcan as a prerequisite for "lights-out" 24/7 operation, in which a deployed fleet must handle its own mechanical emergencies in order to avoid pausing on hardware faults. Because Figure 03 computes joint torque directly from pixels and proprioceptive state through learned networks, the controller can rapidly redistribute load across remaining joints in a way that is difficult to express in hand-written C++.[35]
In parallel with Helix 02, Figure AI announced two related initiatives to scale training data for its VLA. The first is the Helix Lab, a dedicated research and data-collection facility focused on capturing large volumes of egocentric human video and interaction data across real-world environments. The lab integrates data collection, simulation, and on-robot learning into a single pipeline, with the explicit goal of training Helix for deployment on Figure 03.[15]
The second is a strategic partnership with Brookfield Asset Management, announced in September 2025. Brookfield, which manages over $1 trillion in assets and oversees a real estate portfolio that includes approximately 100,000 residential units, more than 500 million square feet of commercial office space, and 160 million square feet of logistics space, is collaborating with Figure to capture human video data across this portfolio. Figure has stated that the collaboration is intended to produce "the world's largest and most diverse real-world humanoid pretraining dataset," and Brookfield also invested in Figure's Series C round as part of the broader partnership.[16]
Figure AI has positioned Figure 03 as the company's first robot designed primarily for home environments. Demonstrations released alongside the October 2025 unveiling showed the robot performing a range of domestic tasks in simulated home settings. These included folding towels and clothing, loading and unloading dishwashers, clearing and wiping tables, sorting and loading laundry, watering plants, serving tea, and navigating between rooms while responding to voice commands.[17][18]
The robot's ability to perform these tasks relies on the combination of Helix's zero-shot generalization (allowing it to handle novel household objects), the fingertip tactile sensors (enabling gentle handling of fragile items), and the palm cameras (allowing manipulation of objects that are partially hidden or in tight spaces such as dishwasher racks). Figure has noted that the four-minute autonomous dishwasher task featured in the Helix 02 launch is a representative example of the long-horizon, multi-step domestic work the platform is being trained to handle without intervention.[3][14]
In March 2026, Figure stated that Figure 03 had demonstrated mastery of eight distinct autonomous cleaning skills during testing, including loading laundry, folding towels, wiping surfaces, and clearing dishes. Figure has also demonstrated Figure 03 doing logistics work, where Helix improved package handling speed to roughly 4.05 seconds per package (down from approximately 5 seconds), an improvement of about 20% within months of initial deployment, while extending its handling to deformable poly bags and flat envelopes alongside rigid boxes.[19]
In the spring of 2026, Figure released two demonstrations that pushed Helix 02 beyond the single-robot, single-task regime of the launch dishwasher video.
On May 8, 2026, Figure published footage of two Figure 03 units resetting a bedroom in under two minutes. The robots performed eight major behaviors in sequence and in parallel, including opening doors, pushing an office chair under a desk, hanging garments on a coat rack, placing headphones on a vertical stand, closing books, operating a pedal-actuated trash can while balancing on one foot, manipulating shared deformable objects, and lifting, unfolding, and smoothing bedding to make the bed. The two robots ran a single learned VLA policy and coordinated their actions without explicit message passing between machines. Each unit perceived the other through its own cameras and inferred its partner's intent from observed motion alone, which Figure described as the first demonstration of a single learned neural network producing collaborative multi-humanoid loco-manipulation directly from pixels.[34]
On May 13, 2026, Figure released a video of a team of Figure 03 units running a full eight-hour autonomous shift on a small-package sorting conveyor belt. The robots picked packages, detected barcodes from camera pixels alone, and reoriented each parcel barcode-down for downstream scanning, processing one package roughly every three seconds at human-comparable throughput. All inference ran on the onboard GPUs with no cloud connectivity, and the eight-hour duration was the longest continuous autonomous shift Figure had publicly demonstrated. CEO Brett Adcock wrote that the robots were "running a full 8-hr shift at human performance levels" and that the run was "fully autonomous running Helix-02." The demonstration extended an earlier one-hour test and was presented as a step toward continuous, lights-out logistics operation.[32]
Figure 03 is not yet capable of fully autonomous, all-day household operation across the full range of consumer scenarios. Independent reviews have noted that the publicly demonstrated home tasks are usually conducted in controlled settings with simple objects, even lighting, and no time pressure. Some tasks (such as folding T-shirts and other unstructured fabrics) remain difficult, and most general-purpose claims about home humanoids in the 2025 to 2026 period have rested on staged demonstrations rather than continuous, in-home autonomy. Initial home deployments are limited to select partner households, and the robot is not yet available for general consumer purchase. Reviewers have also flagged that even the May 2026 multi-robot demonstrations are still relatively short in absolute duration compared to a real domestic day, and that the eight-hour logistics shift was performed in a structured warehouse cell rather than an unstructured home.[20]
To support mass production of Figure 03, Figure AI developed BotQ, a dedicated high-volume humanoid manufacturing facility located on the company's campus in San Jose, California. BotQ occupies an approximately 98,700-square-foot industrial flex building at 3960 North 1st Street, on the Assembly at North First tech campus, which Figure AI leased in 2025 ahead of the Figure 03 launch. The facility represents Figure AI's decision to bring humanoid robot manufacturing fully in-house, controlling the build process, quality assurance, and supply chain rather than relying on contract manufacturers. The company spent approximately eight months developing the manufacturing processes and a further six months building the supporting software infrastructure.[21][36]
The first-generation BotQ production line has an initial capacity of up to 12,000 humanoid robots per year. Figure AI's stated goal is to produce a total of 100,000 robots over four years, supported by a supply chain designed to scale to roughly 3 million actuators in the same period, and the company has publicly discussed an interim annual run-rate target of 50,000 units. The facility employs die-casting, injection molding, metal injection molding, and stamping processes, and vertically integrates the production of actuators, batteries, sensors, and electronics. Figure has noted that parts that previously required more than a week of CNC machining can now be manufactured in under 20 seconds using tooled processes. Figure AI also developed an internal Manufacturing Execution System (MES) to coordinate production across more than 150 networked workstations, alongside in-house product lifecycle management (PLM), enterprise resource planning (ERP), and warehouse management systems (WMS).[3][21]
A distinctive feature of BotQ is the deployment of Figure's own humanoid robots on the assembly line. Once trained on Helix, Figure 03 units assist with material handling between stations and assemble select components of the production line. Figure refers to this as a "robots building robots" approach and has framed it both as a productivity strategy and as a source of training data, since each hour a robot spends on the line generates supervised manipulation experience.[21]
A major achievement in the Figure 02 to Figure 03 transition was a 90% reduction in component costs. The engineering team redesigned nearly every component with manufacturability in mind, shifting from complex CNC machining processes to low-cost, high-volume tooled processes. This cost reduction is central to Figure AI's goal of eventually offering the robot at a consumer-accessible price point below $20,000. Some early third-party coverage speculated that Figure 03 might launch at $35,000, but Figure has consistently described its long-term consumer target as below $20,000, with commercial customers paying a separate per-hour or per-unit rate during the pre-consumer phase.[2][3]
On April 29, 2026, Figure AI reported that more than 350 Figure 03 units had been delivered from BotQ since the start of production, with the production rate rising from one robot per day at launch to one robot per hour, an approximately 24-fold throughput improvement in under 120 days. End-of-line first-pass yield exceeded 80% and continued to improve weekly. The dedicated F.03 battery line achieved a 99.3% first-pass yield, having shipped over 500 packs, while more than 9,000 actuators across more than 10 distinct product variations had been produced. Each unit undergoes more than 80 functional verification tests prior to deployment, and the line includes more than 50 in-process inspection points.[19]
Third-party analysts using Figure's disclosed throughput estimated April 2026 monthly production at approximately 150 units, with the BotQ line on a trajectory that would close the gap to its 12,000-per-year nameplate capacity during the second half of 2026 if yield and uptime continued to improve at the observed rate.[33]
In April 2026, Figure AI and BMW expanded their long-running Spartanburg partnership into what Figure described as the first paid commercial-scale deployment of general-purpose humanoid robots at an industrial customer. The contract covers an initial fleet of 40 Figure 03 units operating across body-shop and assembly-line workstations at BMW's Spartanburg complex in South Carolina, performing parts-handling, sub-assembly placement, and quality-inspection activities that were previously executed by human workers or fixed automation.[19][33]
The pricing model is reported as a per-hour-of-operation arrangement at approximately $25 per robot-operating-hour, structured to undercut equivalent human labor costs while compensating Figure for the deployment and ongoing on-site support. BMW has framed the arrangement as a long-term strategic commitment to humanoid robotics rather than a one-shot pilot, with phased expansion to additional Spartanburg workstations through the remainder of 2026 and 2027 and separate contracts contemplated for the company's German plants. The deployment builds on the Figure 02 fleet that previously logged more than 1,250 operational hours and contributed to production of over 30,000 BMW X3 vehicles, and it provides Helix and Helix 02 with a continuous source of real-world supervised manipulation data.[7][33]
Figure AI's development has been supported by several major funding rounds:
| Round | Date | Amount | Valuation | Key investors |
|---|---|---|---|---|
| Seed | May 2023 | $70 million | Not disclosed | Brett Adcock ($20M personal), Parkway Venture Capital |
| Series B | February 2024 | $675 million | $2.6 billion | Jeff Bezos, Microsoft, NVIDIA, Intel, Amazon, OpenAI |
| Series C | September 2025 | $1+ billion | $39 billion | Parkway Venture Capital, Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Qualcomm Ventures, Salesforce, T-Mobile Ventures, Align Ventures, Tamarack Global, LG Technology Ventures |
The Series C round, announced on September 16, 2025, represented a roughly 15-fold increase in valuation from the February 2024 round just 18 months earlier. Total funding raised across all rounds exceeds $1.9 billion, making Figure AI one of the most well-capitalized pure-play humanoid robotics companies in the world. The Series C also formalized the strategic partnership with Brookfield that supports the Helix pretraining dataset effort.[22][23]
Figure 03 enters an increasingly crowded humanoid robot market. Key competitors include:
| Company | Robot | Focus | Notable features |
|---|---|---|---|
| Tesla | Optimus Gen 3 | Industrial, consumer | 22-DOF hands, manufacturing scale, $20,000 to $30,000 target price |
| 1X Technologies | NEO | Home, consumer | $20,000 outright or $499/month subscription, 22-DOF tendon-driven hands |
| Agility Robotics | Digit | Warehouse, logistics | RoboFab factory, deployed in Amazon and GXO facilities |
| Unitree | G1 / R1 | Consumer, education | R1 priced at $5,900, ultra-lightweight |
| Boston Dynamics | Atlas (electric) | Research, industrial | Advanced mobility and acrobatics |
| Apptronik | Apollo | Industrial, logistics | NASA heritage, modular design |
| Sanctuary AI | Phoenix | General purpose | Carbon (proprietary AI), teleoperation-first approach |
The two robots most directly comparable to Figure 03 in 2025 to 2026 are the 1X NEO and the Tesla Optimus Gen 3. NEO and Figure 03 both target a roughly $20,000 outright price point and explicitly market themselves as home robots, while Tesla Optimus Gen 3 is positioned for industrial deployment first.
| Feature | Figure 03 | 1X NEO | Tesla Optimus Gen 3 |
|---|---|---|---|
| Manufacturer | Figure AI | 1X Technologies | Tesla |
| Height | 1.68 m | ~1.65 m (5 ft 5 in) | 1.73 m |
| Weight | 60 kg | ~30 kg | 57 kg |
| Hand DOF | 16 per hand | 22 per hand (tendon-driven) | 22 per hand |
| Payload | 20 kg | ~25 kg carrying / ~68 kg lifting | Not officially disclosed |
| Battery runtime | ~5 hours | ~4 hours | Not officially disclosed |
| Charging | Wireless inductive (2 kW, foot coils) | Wired | Wired |
| Exterior | Soft washable textiles + foam | Soft knitted fabric | Hard polymer shell |
| AI system | Helix VLA (in-house) | Redwood AI (in-house) | Tesla FSD-derived stack |
| Target price | Below $20,000 (consumer target) | $20,000 outright or $499/month | $20,000 to $30,000 (estimated) |
| Availability (mid-2026) | Commercial pilots; partner-only home | US consumer pre-orders open; deliveries through 2026 | Tesla internal use; external customers later |
| Primary market focus | Home + commercial | Home (with teleop fallback) | Industrial first |
At around 60 kilograms, Figure 03 is approximately twice the mass of 1X NEO, reflecting two different home-safety philosophies. Figure has invested in a soft textile cover, fingertip tactile sensing, and the Vulcan balance policy to make a heavier, more capable platform behave safely around people. 1X has instead made a lighter, gentler platform a primary feature, accepting reduced payload and slower mechanical bandwidth to make the robot easier to catch if it falls and less destructive on impact. NEO also retains a hybrid teleoperation fallback in which remote operators can take control when the autonomous policy fails, a model Figure has so far declined to adopt for Figure 03 home deployments.[24][25]
Figure 03 differentiates itself through its home-first hardware design (soft textiles, wireless charging, safety-focused exterior), its proprietary vertically integrated AI stack (Helix and Helix 02), and its in-house manufacturing capability (BotQ). NEO competes on a similar consumer thesis but with a much lighter chassis and a hybrid teleoperation fallback, while Tesla Optimus prioritizes deployment within Tesla's own factories before targeting external customers.[24][25]
The global humanoid robot market was valued at approximately $4.89 billion in 2025 and is projected to grow to $165 billion by 2034, reflecting a compound annual growth rate of more than 50%. Chinese manufacturers, particularly Unitree and AgiBot, led global shipments in 2025 with over 10,000 robots combined, while U.S. companies including Tesla, Figure AI, and Agility Robotics each shipped roughly 150 units, underscoring the early stage of Western market development. Figure AI's BotQ ramp through April 2026 indicated that the next year may significantly reshape the U.S. share of global humanoid shipments.[26]
Figure 03 was named one of TIME's Best Inventions of 2025 and appeared on the cover of the magazine's annual Best Inventions issue, accompanied by an exclusive cover story by journalist Billy Perrigo profiling Figure AI's ambitions to create a mass-producible home humanoid. The article noted that Figure 03's joints are smaller and stronger than those of Figure 02, that its components cost roughly 90% less to manufacture, that its hands are slimmer with tactile finger pads and a palm camera, and that its battery is less prone to thermal runaway than earlier designs. The 300-invention list in 2025 was the largest TIME has ever published. The Unitree R1 was also among the humanoid robots named to the list, reflecting the broader acceleration of humanoid hardware in that year.[1][27]
In March 2026, Figure 03 appeared at a White House event hosted by First Lady Melania Trump on March 25. The two-day summit, titled "Fostering the Future Together Global Coalition Summit," focused on AI in education and gathered first spouses from 45 nations. Figure AI described it as the first humanoid robot to appear inside the White House. Figure 03 walked alongside Mrs. Trump along a red carpet in the East Room, greeted attendees, said "I'm Figure 03, a humanoid built for the United States of America," and delivered short greetings in ten languages. CEO Brett Adcock posted that he was "proud to see F.03 make history as the first humanoid robot in the White House." The event generated significant global media coverage and public discussion about the role of humanoid robots in education and daily life.[28][29]
In November 2025, Figure AI was sued by Robert Gruendel, the company's former head of product safety. The lawsuit, reported by CNBC and other outlets on November 21, 2025, alleged that internal impact testing of the Figure 02 had generated forces more than double the level required to fracture an adult human skull, and that the company's next-generation robots moved at superhuman speeds that introduced significant safety risk. As evidence, the complaint cited an internal incident in which a malfunctioning robot allegedly left a three-quarter-inch gash in a stainless-steel refrigerator door. Gruendel said he was terminated in September 2025, days after lodging detailed safety complaints, and alleged that executives had "gutted" his safety roadmap to accelerate deployment ahead of a funding round. He sought economic, compensatory, and punitive damages and a jury trial. Figure AI denied the claims, attributing Gruendel's termination to "poor performance" and stating that it would "thoroughly discredit" the allegations in court. The lawsuit prompted broader discussion of how humanoid robot makers should approach safety as their machines move from controlled factory cells into homes.[30][31]
| Feature | Figure 01 (2023) | Figure 02 (2024) | Figure 03 (2025) |
|---|---|---|---|
| Manufacturer | Figure AI | Figure AI | Figure AI |
| Height | 1.68 m | 1.68 m | 1.68 m |
| Weight | 60 kg | ~70 kg | 60 kg |
| Total DOF | Not disclosed | 41 | 44 |
| Hand DOF | Not disclosed | 16 | 16 |
| Payload | 20 kg | 25 kg | 20 kg |
| Battery runtime | ~5 hours | ~5 hours | ~5 hours (F.03, 2.3 kWh) |
| Battery placement | External backpack | Torso-integrated | Torso-integrated, structural |
| Cameras | Cameras + LiDAR | 6 RGB | 8 (6 main + 2 palm) |
| Tactile sensors | Basic | Force sensors | 3-gram fingertip sensors |
| Charging | Wired | Wired | Wireless inductive (2 kW) |
| Exterior | Hard shell, external cabling | Hard shell, internal cabling | Soft washable textiles + foam |
| AI system | None (teleoperated) | Onboard VLA + OpenAI LLM | Helix VLA (in-house); Helix 02 (Jan 2026); Vulcan (Apr 2026) |
| Compute | Not disclosed | 3x Figure 01 | Dual embedded GPUs |
| Manufacturing | Prototype | Hand-built / pilot | BotQ (12,000 / yr line; 1 robot / hr by Apr 2026) |
| Target use | Research / warehouse | Industrial (BMW) | Home + commercial |