Tag: Figure AI

  • Factory and Manufacturing Robots and Drones in 2026: Inside the World’s Largest, Oldest, and Most Operationally Mature Robotics Deployment

    In November 2025, a California-based humanoid robotics company called Figure AI announced the official retirement of its Figure 02 humanoid platform after the completion of an 11-month pilot deployment at BMW Manufacturing’s Spartanburg, South Carolina assembly plant. The operational data Figure published with the retirement announcement was the most detailed disclosure ever made of a humanoid robot’s performance in an active commercial factory. Two Figure 02 units, each 170 centimeters tall, 70 kilograms in mass, with a 20-kilogram payload capacity, operated 10-hour shifts Monday through Friday on the BMW X3 body shop line, performing the specific operational task of removing sheet-metal parts from racks and bins and placing them onto welding fixtures with a 5-millimeter tolerance, on an 84-second cycle time (37 seconds for the load alone). The robots accumulated 1,250 hours of runtime, loaded more than 90,000 sheet-metal parts, contributed to the production of more than 30,000 BMW X3 vehicles, walked approximately 1.2 million steps covering an estimated 200 miles inside the plant, and maintained placement accuracy above 99 percent across the deployment. Brett Adcock, Figure’s CEO, accompanied the retirement announcement with photos of the Figure 02 units returning to Figure’s headquarters covered in scratches, scuffs, and industrial grime. The forearm subsystem, by Figure’s own disclosure, was the top hardware failure point. The lessons would, by Figure’s stated plan, be integrated into the next-generation Figure 03 platform launching for production deployment in 2026.

    The BMW Spartanburg deployment is, in 2026 operational terms, the most heavily-documented humanoid-robot-in-factory deployment in the commercial history of industrial robotics. It is also, by every available measure of deployed-unit count, an almost negligible fraction of the actual industrial robotics installed base operating inside the world’s factories in 2026. The International Federation of Robotics estimates the global industrial robot installed base passed 4 million units in 2024 — bolted-down articulated arms, SCARA robots, parallel-kinematic delta robots, and collaborative robots operating in continuous production across the automotive, electronics, metals, plastics, food-and-beverage, pharmaceutical, and aerospace manufacturing sectors. The first industrial robot — Unimate, designed by George Devol and Joseph Engelberger — was installed at a General Motors plant in Trenton, New Jersey in 1961. The factory robotics industry has 65 years of operational deployment behind it. The humanoid robot pilots at BMW, Mercedes-Benz Berlin-Marienfelde, Tesla Fremont, GXO Logistics Atlanta, and the growing list of automotive and logistics factory pilots are, in installed-base terms, a few hundred units against an installed base of 4 million conventional industrial robots that have been quietly producing the physical objects of the modern economy since before most of the people designing humanoid robots were born.

    The Big Four industrial robot manufacturers

    The global industrial robotics market is dominated, by both installed base and annual installations, by four manufacturers: FANUC Corporation (Japan), ABB Group (Switzerland), KUKA AG (Germany), and Yaskawa Electric Corporation (Japan). FANUC, headquartered at the foot of Mount Fuji in Oshino-mura, Yamanashi Prefecture, builds the yellow-painted articulated robots that have become the visual signature of automotive paint shops, electronics assembly lines, and metal-fabrication facilities globally. FANUC’s installed base is approximately 750,000 deployed industrial robots worldwide, with the M-410, R-2000iC, LR Mate, and CRX collaborative robot product lines spanning payload capacities from 4 kilograms (LR Mate) to 2,300 kilograms (M-2000iA, the company’s heaviest articulated arm). FANUC also manufactures the Roboshot injection-molding machines, the Robocut wire EDM machines, and the Robodrill small-machining centers — the broader factory automation product line that has, in operational terms, made FANUC one of the most consistently profitable Japanese industrial conglomerates over the past three decades.

    ABB Group, headquartered in Zurich, builds the IRB series of articulated robots and the YuMi dual-arm collaborative robot, with installed-base estimates in the 500,000-600,000 unit range globally. ABB’s industrial automation business operates across the same automotive, electronics, food-and-beverage, and metals manufacturing segments as FANUC, with particular strength in European automotive deployment. KUKA AG, headquartered in Augsburg, Germany, builds the orange-painted KR series of articulated robots that has been operationally synonymous with German automotive manufacturing for decades — KUKA robots populate the assembly lines at Volkswagen, BMW, Mercedes-Benz, and Audi facilities across Europe at deployment volumes no Japanese or American manufacturer approaches. KUKA was acquired by Midea Group — the Chinese consumer appliance conglomerate — in a 2017 transaction that, despite the substantial geopolitical attention it received at the time, has produced relatively continuous operational management since the transaction closed. Yaskawa Electric, the Kitakyushu-based Japanese manufacturer, operates the Motoman robot brand, with the GP, MH, and AR series spanning the standard industrial-robot payload range and an installed base in the 500,000-plus unit range.

    The Big Four collectively account for, by industry analyst estimates, approximately 55 to 65 percent of global industrial robot installations in any given year. The remaining 35 to 45 percent is distributed across a long tail of specialist manufacturers — Kawasaki Heavy Industries, Mitsubishi Electric (Melfa series), Denso Corporation (VS series), Stäubli (TX and TS series), Epson Robots (SCARA platforms), Nachi-Fujikoshi, and increasingly the Chinese manufacturers discussed below. The product taxonomy of conventional industrial robots is highly standardized across these manufacturers: six-axis articulated robots for general assembly, SCARA robots for high-speed pick-and-place, delta robots for high-throughput packaging, palletizing robots for warehouse end-of-line operations, and collaborative robots (cobots) for human-robot shared workspace applications. The form factors, control architectures, and operational deployment patterns have, over the past 30 years, converged on a set of standards that the entire factory automation industry operates against.

    The cobot category: Universal Robots, Doosan, Techman, and the small-payload collaborative wave

    The fastest-growing subcategory within industrial robotics over the past decade has been collaborative robotics — the smaller, force-limited, vision-aware articulated arms designed to operate alongside human workers without traditional safety cages or perimeter fencing. The category-leading manufacturer is Universal Robots, the Danish company founded in 2005 in Odense and acquired by Teradyne (NASDAQ:TER) in 2015 for approximately $285 million. Universal Robots has, as of 2024, deployed more than 75,000 cobots globally across the UR3, UR5, UR10, UR16, and UR20 product lines, with the UR15 platform launching in March 2025 as the company’s most recent product addition. The Universal Robots cobot architecture — a six-axis articulated arm with force-torque sensing at every joint, a polycarbonate enclosure, payloads ranging from 3 kilograms (UR3) to 30 kilograms (UR30), and a unified control architecture that enables relatively rapid task programming compared to traditional industrial robots — has become the dominant operational template for the broader cobot category.

    The competing cobot manufacturers include Techman Robot (Taiwan, owned by Quanta Computer since 2018, builder of the TM series cobots with integrated machine vision), Doosan Robotics (South Korea, the M and H series cobots, IPO’d on the Korea Exchange in October 2023), Franka Emika (Munich-based, the Panda cobot platform, restructured under bankruptcy in 2023 and acquired by Cologne-based industrial robotics company Agile Robots SE), AUBO Robotics (Chinese-American joint venture), Productive Robotics (U.S.-based OB7 cobot), and the cobot lines from the Big Four (FANUC CRX, ABB YuMi and GoFa, KUKA LBR iiwa, Yaskawa HC-series). The cobot market in 2026 is estimated at approximately $2.5 billion in annual revenue, with double-digit annual growth rates substantially exceeding the broader industrial robotics market’s mid-single-digit growth.

    The Chinese industrial robotics rise: Estun, Inovance, EFORT, and the Made in China 2025 acceleration

    The single most operationally consequential shift in factory robotics over the 2020-2026 window has been the rise of Chinese industrial robot manufacturers. China became the world’s largest annual industrial robot market by installations in approximately 2016 and has, by IFR data, accounted for approximately 52 percent of global industrial robot installations in 2024 — more than 290,000 newly-installed robots in China alone against a global total of roughly 560,000 installations. The shift on the demand side was followed by an equally significant shift on the supply side. Estun Automation (Nanjing, Shenzhen-listed under 002747.SZ), Inovance Technology (Shenzhen, listed under 300124.SZ), EFORT Intelligent Equipment (Wuhu, listed under 688165.SH), Siasun Robot & Automation, STEP Electric Corporation, and Han’s Robot have, over the 2018-2026 window, collectively grown from minor domestic players to genuine global competitors. Estun, in particular, has emerged as the largest Chinese industrial robot manufacturer by deployed units, with an installed base in the 100,000-plus range as of 2024 and acquisitions across the European industrial automation supply chain — including the 2017 acquisition of TRIO Motion Technology in the United Kingdom and the 2019 acquisition of German automation specialist Cloos Schweißtechnik.

    The structural driver behind the Chinese industrial robotics rise is the Made in China 2025 industrial policy, launched in 2015 by the Chinese State Council, that designated industrial robotics as one of ten priority strategic sectors for domestic capability development. Combined with the broader dual-circulation economic strategy announced in 2020, the policy framework has funneled substantial state-directed investment into Chinese industrial robotics manufacturers, robotic component suppliers (precision reducers, servo motors, controller electronics), and downstream factory automation deployment across Chinese manufacturing. The 2024-2026 acceleration has been driven by the broader decoupling pressures between Chinese manufacturing and Western technology supply chains, with Chinese manufacturers increasingly required by state-directed procurement policies to source domestic industrial automation equipment where viable.

    The humanoid robot factory wave: Figure, Tesla Optimus, Apptronik, Agility, and the auto-and-logistics pilot deployment cohort

    The humanoid robot wave that began commercial factory pilot deployment over the 2023-2026 window is, in operational terms, the most heavily-financed and most-publicized but smallest-by-deployed-unit-count segment of the broader factory robotics market. The Figure 02 BMW Spartanburg pilot is the most operationally documented example. Tesla‘s Optimus platform has been deployed inside Tesla’s Fremont, California and Austin, Texas vehicle manufacturing facilities for testing and routine task execution, with Elon Musk publicly stating in multiple 2024-2025 earnings calls that Tesla is targeting thousands of Optimus units in internal factory deployment by 2026. Apptronik‘s Apollo platform has been deployed at Mercedes-Benz manufacturing facilities in Berlin-Marienfelde and Kecskemét, Hungary, and inside Jabil electronics-manufacturing operations under the strategic partnership announced in February 2025. Agility Robotics‘ Digit has been deployed at GXO Logistics Spanx fulfillment operations in Atlanta and at additional logistics customer sites. 1X Technologies‘ Neo has been deployed in pilot facilities, with the company having raised more than $100 million from investors including OpenAI. Hexagon Robotics‘ AEON humanoid, unveiled in June 2025, began pilot deployment at BMW’s Leipzig plant in December 2025 as the second humanoid robot deployed within the BMW iFACTORY initiative, alongside the broader Boston Dynamics Spot quadruped fleet that has been operating in BMW and Hyundai factory inspection routines since 2022. Foxconn has, since 2023, publicly disclosed development of humanoid robotics in partnership with NVIDIA’s Project GR00T platform for deployment in its electronics-manufacturing operations, with the underlying foundation-model work increasingly conducted in collaboration with academic robotics research labs at Stanford, MIT, Carnegie Mellon, and UT Austin.

    The structural observation about the humanoid factory wave in 2026 is that the total deployed unit count across all manufacturers globally is, by available public disclosure, in the low thousands — roughly 0.05 to 0.1 percent of the broader industrial-robot installed base. The pilots are operationally important. The Figure 02 BMW deployment has generated more public-facing data about humanoid factory performance than any prior deployment. The Tesla Optimus internal deployments — though Tesla has disclosed less specific operational data than Figure has — have, by Musk’s public claims, achieved meaningful internal factory utility. But the bolted-down FANUC, ABB, KUKA, and Yaskawa industrial robots that have populated the world’s factories for 60 years continue to outnumber the humanoid platforms by approximately 1,000 to 1 in deployed-unit terms, and continue to perform the bulk of the actual manufacturing work in the global economy in 2026.

    Robot density: South Korea, Singapore, Germany, Japan, and the international competitiveness story

    The most useful single statistic for understanding the international competitive dynamics of factory automation is robot density — the number of operational industrial robots per 10,000 manufacturing workers in a given economy. IFR data for 2022-2023 placed South Korea at approximately 1,012 robots per 10,000 manufacturing workers — the highest robot density in any major economy in the world by a significant margin. Singapore was second at approximately 770. Germany was third at approximately 415. Japan was fourth at approximately 397. China had climbed to fifth place at approximately 322 robots per 10,000 manufacturing workers — a substantial increase from sub-100 a decade earlier. The United States was sixth at approximately 285, with Sweden, Denmark, Hong Kong, and Taiwan rounding out the top ten. The implication for U.S. manufacturing competitiveness is direct: South Korea operates approximately 3.5 times more industrial robots per manufacturing worker than the U.S. does, and the gap has been widening since approximately 2018 rather than narrowing.

    The structural driver behind the South Korean robot-density lead is the heavy concentration of South Korean manufacturing in two sectors — automotive (Hyundai, Kia, KG Mobility) and electronics (Samsung, LG, SK Hynix) — both of which are extremely high-automation industries by global standards, and both of which have been actively automating since the 1990s under coordinated industrial policy. The structural driver behind the Singapore robot-density figure is the electronics manufacturing concentration in the Singaporean economy combined with active state-led automation incentives. The structural driver behind the German robot-density is the legacy of German automotive manufacturing’s longstanding automation leadership and the broader Mittelstand mechanical-engineering ecosystem. The structural driver behind the U.S. relative lag is harder to summarize cleanly — the U.S. manufacturing sector is more heterogeneous (broader range of industries), the labor cost gap between manual labor and automation has been smaller for most of the past 30 years than in higher-cost economies, and the historical U.S. manufacturing offshoring wave to Mexico, China, and Southeast Asia reduced the demand for domestic factory automation through the 2000s and 2010s.

    The reshoring wave and the CHIPS Act / IRA / IIJA buildout context

    The single largest demand-side accelerator for U.S. factory robotics in the 2024-2026 window has been the convergence of three federal industrial-policy initiatives: the CHIPS and Science Act (signed August 2022, authorizing approximately $52 billion in semiconductor manufacturing incentives), the Inflation Reduction Act (August 2022, approximately $369 billion in clean energy spending including electric vehicle and battery manufacturing incentives), and the Infrastructure Investment and Jobs Act (November 2021, $1.2 trillion in infrastructure spending). The CHIPS Act has driven major semiconductor manufacturing facility construction at TSMC Arizona (Phoenix), Intel Ohio (New Albany), Samsung Texas (Taylor), Micron New York (Syracuse), and GlobalFoundries New York (Malta). The IRA has driven major battery and EV manufacturing facility buildouts at Tesla Gigafactory Nevada (expansion), Tesla Gigafactory Texas (Austin), Hyundai Metaplant (Bryan County, Georgia), Ford BlueOval City (Tennessee), Volkswagen Scout Motors (South Carolina), and LG Energy Solution, SK Innovation, Panasonic, and CATL battery manufacturing investments across multiple U.S. states. Each of these new facilities represents tens of thousands of square feet of greenfield factory floor space requiring industrial robotics deployment from initial buildout, and each represents capital deployment that conventional manufacturing-equipment depreciation cycles would otherwise have spread across decades.

    The structural reshoring trend has, by every available measure, been the most consequential single demand driver for U.S. factory automation since the 1990s. The factories being built are being built with substantially higher automation densities than the U.S. manufacturing facilities they are notionally replacing, in part because the labor cost equation no longer supports manual-labor-intensive operations at U.S. wage levels and in part because the semiconductor and battery manufacturing processes being deployed are inherently more automation-dependent than the consumer electronics and automotive operations that previous waves of U.S. manufacturing offshored.

    The factory drone category: Verity, Pinc Solutions, and the indoor inventory inspection niche

    The drone category in factory operations is, in operational terms, much smaller than the industrial-robot category, but it occupies a specific niche around indoor inventory inspection and asset surveillance. Verity AG, the Zurich-based industrial drone company, builds fully-autonomous indoor drones that operate inside warehouses and distribution centers, scanning RFID-tagged inventory pallets, capturing visual documentation of stock positions, and feeding data into warehouse-management systems. Verity has deployed across Nestlé, Maersk, DSV, and Geodis warehouse operations. Pinc Solutions operates a competing indoor inventory drone platform deployed at Ralph Lauren, Lego, and Bridgestone distribution facilities. Eyesee (a subsidiary of Hardis Group, France) operates the Eyesee indoor warehouse inventory drone. The indoor warehouse drone category, while smaller in revenue than the broader industrial-robot category, has demonstrated the operational use case for autonomous aerial robotics in structured indoor environments where the outdoor drone navigation challenges do not apply.

    The outdoor factory drone category — perimeter security, smokestack and refinery inspection, solar array inspection, large facility surveying — is dominated by the same drone manufacturers serving construction and infrastructure inspection markets: DJI (Phantom 4 RTK, Matrice 350 RTK, Mavic 3 Enterprise), Skydio, Parrot Anafi USA, and Flyability‘s Elios confined-space inspection drone, which operates inside boilers, storage tanks, and other enclosed industrial spaces.

    What 2026 looks like across factory and manufacturing robotics

    In 2026, the factory robotics category is structurally dominated by the conventional industrial robot installed base — approximately 4 million deployed units globally, growing by 500,000-plus annual installations, dominated by FANUC, ABB, KUKA, and Yaskawa with the long tail of specialist manufacturers and the rapidly-growing Chinese manufacturers (Estun, Inovance, EFORT) accounting for the balance. The cobot category, dominated by Universal Robots with Techman, Doosan, and the Big Four’s cobot lines competing, continues to be the fastest-growing subcategory at approximately $2.5 billion in annual revenue. The humanoid factory wave — Figure (post-02 retirement, transitioning to Figure 03), Tesla Optimus, Apptronik Apollo (Mercedes-Benz, Jabil), Agility Digit (GXO, Amazon), 1X Neo, Hexagon AEON (BMW Leipzig), and the Foxconn-NVIDIA humanoid manufacturing initiative — operates at deployed-unit volumes in the low thousands against the four-million-unit conventional installed base, with the Figure 02 BMW Spartanburg deployment standing as the most operationally documented humanoid-in-factory deployment in commercial history. South Korea operates at 1,012 robots per 10,000 manufacturing workers; the U.S. operates at 285. The CHIPS Act, IRA, and IIJA federal industrial policy is driving the largest U.S. factory buildout in three decades, with TSMC Arizona, Intel Ohio, Samsung Texas, and the broader EV-and-battery manufacturing investment wave creating the demand environment for accelerated factory automation deployment.

    The structural story across factory robotics in 2026 is that the category is, simultaneously, the most operationally mature and the most actively disrupted of any robotics deployment domain. The bolted-down industrial robot has 65 years of operational deployment behind it — six decades that no other robotics category approaches. The 4 million installed units perform the bulk of the actual manufacturing work in the global economy and will continue to do so for the operational lifetime of the equipment currently deployed. But the category is also being actively disrupted on multiple vectors simultaneously: Chinese manufacturers competing with the historical Big Four on cost and increasingly on capability, cobot manufacturers expanding the addressable market into smaller manufacturers that conventional industrial robots could not serve, humanoid robot manufacturers piloting platforms that — if the operational reliability projected by Figure, Tesla, Apptronik, Agility, and 1X actually materializes at scale — could expand the addressable factory-automation market by an order of magnitude over the 2026-2035 window. The category is dominated by mature platforms doing routine work, layered over by a small number of high-attention-receiving experimental platforms that may or may not eventually justify the venture capital and corporate-strategic investment they have received.

    The Figure 02 BMW deployment is the operational data point that defines what the answer might look like. Eleven months. 1,250 hours. 90,000 sheet-metal parts. 30,000 BMW X3 vehicles. 99 percent placement accuracy. A forearm subsystem that emerged as the top hardware failure point — and a Figure 03 platform launching in 2026 that will, by Figure’s stated plan, address the specific hardware reliability lessons learned at Spartanburg. The traditional six-axis FANUC welding robot down the line that received the sheet-metal parts the Figure 02 robots loaded did not generate a press release. The traditional robot has been doing that exact task in some configuration since approximately 1985. The traditional robot is the deployed industrial economy. The humanoid platform is the deployment experiment that, depending on how the Figure 03 / Optimus / Apollo / Digit / AEON / Neo cohort performs over the 2026-2030 window, could either become the next mature deployment template or could remain a high-visibility experimental category that the conventional industrial-robot installed base ultimately absorbs without fundamental architectural change.

    The data that will resolve that question over the next five years is being generated, in 2026, inside the same global factory installed base that has been quietly producing the physical objects of the modern economy for six decades. The robots that move the global trade flows, patrol oil-and-gas facilities, deliver hospital prescriptions to patient homes, retrofit excavators into autonomous solar pile drivers, respond to wildfires and structural collapses, scout planetary surfaces beyond Earth, count penguins in Antarctica, and throw 100-mph cutters in MLB clubhouses all derive, in mechanical engineering, control architecture, and operational deployment terms, from the bolted-down industrial robot that George Devol and Joseph Engelberger installed at the General Motors Trenton plant in 1961. The factory is the parent industry. Everything else is a derivative deployment of the operational principles that the factory automation industry has been refining since the Eisenhower administration. The robots that work at scale in 2026 — anywhere in the economy, in any application — work because the conventional industrial-robot industry figured out, six decades ago, that automation is not about replacing humans wholesale but about deploying specialized machines for specific repetitive tasks under operational constraints that the broader industrial supply chain can actually sustain. The Figure 02 BMW pilot is, in operational terms, the same kind of deployment experiment that General Motors ran with Unimate in 1961. The result, after sixty-five years of cumulative learning, is the 4-million-unit global installed base that quietly produces almost everything else.

    The next sixty-five years will be either an extension of that operational logic into humanoid-robot territory or a continuation of the bolted-down articulated-arm dominance that has, on the available evidence, been the most successful single deployment template in the history of industrial automation. Which of those two outcomes materializes depends on a small number of specific operational variables — humanoid hardware reliability at scale, the training of the next generation of robotics engineers, the comparative cost trajectories of humanoid versus conventional platforms — that are being actively worked on inside Figure, Tesla, Apptronik, Agility, 1X, FANUC, ABB, KUKA, Yaskawa, Estun, and the broader factory robotics industry in 2026. The answer is not yet known. The deployment data being generated in the meantime, including the Figure 02 / BMW Spartanburg pilot, is what will eventually determine which template wins.

  • Boston Dynamics vs. Tesla vs. Figure: The Humanoid Robot Race in 2026

    At CES 2026, Boston Dynamics unveiled the production version of Atlas—fully electric, 56 degrees of freedom, 50-kilogram lift capacity, autonomous battery swap—and won CNET’s “Best Robot” award. Every 2026 unit is already committed: they’re shipping to Hyundai’s Robotics Metaplant Application Center and Google DeepMind, with additional commercial customers planned from 2027. Korean securities firms valued Boston Dynamics between $21 and $28 billion, with bullish IPO projections reaching $88 to $103 billion. The company announced a strategic AI partnership with Google DeepMind and Toyota Research Institute. Its outgoing CEO, Robert Playter, said the goal is for Atlas robots to be “contextually aware of their environment and able to use their hands to manipulate any object.”

    That same month, Elon Musk announced that Optimus would go to the Moon. The previous year, he’d said it would go to Mars. Before that, he’d said Tesla would produce 5,000 to 10,000 Optimus units in 2025. The actual production number was reportedly in the hundreds. As of Q1 2026, Tesla confirmed that Optimus was still in an “R&D and learning phase” with no robots performing productive tasks in Tesla factories. The Optimus program lead since 2022, Milan Kovac, resigned in June 2025.

    Meanwhile, Figure AI closed a Series C round valuing the company at $39 billion—for a startup with only a few hundred commercial units deployed. Global robotics investment surpassed $10 billion in 2025. And the gap between valuations, promises, and actual robots doing actual work in actual facilities has never been wider.

    Three philosophies, one question

    The humanoid robot race in 2026 is not a single competition. It’s three companies making fundamentally different bets about what matters most.

    Boston Dynamics is betting on capability first, scale second. Atlas is the culmination of 13 years of continuous development, originally funded by DARPA for search-and-rescue operations. The old hydraulic Atlas could do backflips and run parkour courses. The new electric Atlas retains that dynamic agility while adding the manufacturability and reliability required for commercial deployment. Hyundai, Boston Dynamics’ majority shareholder, has committed $26 billion to U.S. manufacturing that includes a robotics factory capable of producing 30,000 units per year. The estimated price per unit is $140,000 to $150,000—enterprise-grade pricing for enterprise-grade performance. The target customer is a Fortune 500 manufacturer, not a consumer.

    Tesla is betting on scale first, capability second. Optimus is designed from the outset for mass production, leveraging Tesla’s automotive supply chain, manufacturing expertise, and AI infrastructure (the same Full Self-Driving platform that powers its vehicles). Musk’s target price is $20,000 to $30,000—deliberately “less than a car.” At that price point, if the robot can perform useful tasks, the addressable market is essentially every warehouse, factory, and eventually every household on earth. The problem is the “if.” Every public Optimus demonstration has been criticized for signs of remote human control. Tesla has never held a fully autonomous public demonstration without controversy. The V2.5 iteration, revealed in late 2025, improved the cosmetic design but reviewers described it as underwhelming in function—slow voice command response, tentative motion, awkward pauses. Cosmetic refinement outpacing demonstrable capability is not the trajectory you want if your thesis depends on the robot actually working.

    Figure AI is betting on speed and capital. Founded in 2022, the company raised over $1.6 billion and reached a $39 billion valuation in roughly three years—a pace that would be remarkable even by Silicon Valley standards. Figure’s approach combines a hardware platform (the Figure 02, estimated at over $100,000 per unit) with aggressive AI integration, including a partnership with OpenAI for natural language interaction and task understanding. The company has deployed a few hundred units commercially and is positioning itself as the startup most likely to bridge the gap between research platform and scalable product. Whether a three-year-old company can compete with Boston Dynamics’ 30 years of locomotion research and Tesla’s manufacturing infrastructure is the open question, but the capital markets are clearly betting that the AI component—making the robot understand what you want it to do—matters more than the hardware component.

    What the robots can actually do right now

    This is where the marketing and the engineering diverge most sharply.

    Atlas can walk, run, jump, recover from dynamic perturbations, manipulate objects with precision, and navigate unstructured environments. The CES 2026 demonstration showed car part sequencing and factory component handling. It was remotely operated during the demo, but Boston Dynamics stated the commercial version will be fully autonomous. The company has a deployment track record with Spot (the quadruped) and Stretch (the warehouse robot) that provides credibility for its commercialization claims. Atlas’s 56 degrees of freedom, water resistance, and extreme temperature tolerance make it the most physically capable humanoid robot in production.

    Optimus can walk with an improved heel-to-toe gait, perform basic pick-and-place operations, and handle simple household tasks in staged demonstrations—stirring a pot, sweeping, vacuuming. These are legitimate capabilities, but they’re a long distance from productive factory work. Tesla has not published operational metrics like cycle times, task completion rates, or hours of autonomous operation. Independent reporting suggests the robots deployed internally at Tesla factories are in a learning phase rather than performing useful labor. Musk acknowledged in Tesla’s Q4 2025 earnings that the robots are “not doing useful work” yet. Consumer sales are targeted for late 2027.

    Figure 02 has demonstrated warehouse picking and packing tasks, object manipulation, and natural language interaction through its OpenAI integration. The company has deployed units at BMW’s Spartanburg manufacturing plant and other commercial sites. The demonstrations are impressive but limited in scope, and the deployed fleet numbers in the hundreds—enough for pilot programs, not enough for operational conclusions about reliability or economics.

    The China factor

    The comparison that the American companies probably don’t want you making is with Chinese humanoid robot manufacturers, who are approaching the problem from a different angle entirely. Unitree’s humanoid models emphasize agile maneuvers at accessible price points. UBTECH’s Walker series has demonstrated autonomous battery swapping for continuous 24/7 operation—a practical advantage in factory settings where uptime matters more than acrobatics. BYD, the EV manufacturer, targeted 1,500 humanoid robot units in 2025 and is scaling to 20,000 by 2026.

    Chinese firms are pursuing narrow, production-centric optimization: rapid iteration, manufacturability, duty-cycle engineering, and lower unit costs. Their emphasis is less on demonstrating a spectacular generalist and more on producing reliable, maintainable machines for specific operational roles. That approach—boring, incremental, manufacturing-focused—is exactly what scaled industrial deployment actually requires, and it’s the approach most likely to produce the first humanoid robot that earns its keep on a factory floor without a press release attached to every shift.

    Manufacturing costs across the industry dropped roughly 40 percent from 2023 to 2024, falling from $50,000–$250,000 per unit to $30,000–$150,000, according to Goldman Sachs. That trajectory, if it continues, brings the economics of humanoid robots into range for mainstream industrial deployment by the late 2020s regardless of which specific company gets there first.

    The honest scorecard

    On technical capability in 2026: Atlas wins. It’s not close. Thirty years of locomotion research, DARPA funding, the transition from hydraulic to electric, Google DeepMind AI integration, and a parent company willing to build a 30,000-unit factory give it advantages that no competitor can replicate in the short term.

    On manufacturing potential: Tesla wins, theoretically. No company on earth has more experience producing complex electromechanical systems at scale. If Optimus reaches the point where it can perform useful work autonomously, Tesla’s production infrastructure is unmatched. The problem is that “if,” and the gap between Musk’s production targets and actual output has been growing, not shrinking.

    On capital and velocity: Figure AI wins. A $39 billion valuation and aggressive AI partnerships give it the resources and talent to move fast. Whether moving fast in robotics translates to a durable product—as opposed to a series of impressive demos—is unproven.

    On realistic near-term deployment: Boston Dynamics wins again. It’s the only company in this comparison with a production robot shipping to commercial customers in 2026, backed by a parent company with a $26 billion manufacturing commitment and a track record of commercializing previous robots (Spot, Stretch) that actually operate in the field.

    The humanoid robot that reshapes industry probably isn’t the one that does the best backflip or gets the highest valuation. It’s the one that shows up to work on a random Tuesday, completes its task list without human intervention, and does it again the next day for less than the cost of hiring a person. None of the three companies in this comparison has demonstrated that yet. The race isn’t over. It arguably hasn’t started—because the real competition begins when the robots stop performing and start producing.

    We cover the engineering, AI integration, and commercial deployment of humanoid robots across 24 lectures in our Humanoid Robots & Drones course—including why the company that wins the humanoid robot race might not be the one making the best robot.

  • The Uncanny Valley Problem: Why Humanoid Robots Are So Unsettling (And Whether It Matters)

    In January 2026, a team at Columbia Engineering published a paper in Science Robotics announcing that they’d built a robot capable of learning realistic lip movements by watching its own reflection and studying human videos. The robot could speak in multiple languages and sing. Hod Lipson, the lab’s director, framed the significance plainly: “There is no future where all these humanoid robots don’t have a face. And when they finally have a face, they will need to move their eyes and lips properly, or they will forever remain uncanny.” His co-researcher Yuhang Hu added: “We are close to crossing the uncanny valley.”

    That phrase—uncanny valley—has been floating around robotics and cognitive science for over fifty years, and it describes a problem that gets more commercially urgent every quarter. Tesla, Figure AI, Agility Robotics, Boston Dynamics, and a growing roster of companies are building humanoid robots designed to operate in spaces built for humans—warehouses, hospitals, factories, homes. The machines are getting better at walking, grasping, navigating, and following instructions. The question of whether people will actually want to be around them is a different engineering challenge entirely, and it’s one that can’t be solved with better actuators.

    The graph that launched a thousand nightmares

    Japanese roboticist Masahiro Mori proposed the concept in 1970 in an essay for the journal Energy. The idea is deceptively simple: as a robot becomes more human-like in appearance, people’s emotional response becomes more positive—up to a point. A cartoon robot is charming. A robot with a humanoid shape and some facial features is engaging. But as the resemblance approaches near-human levels without quite getting there, the response doesn’t just plateau. It collapses. The emotional curve drops into a trough of revulsion, unease, and what Mori called bukimi—a Japanese word that translates roughly to “eeriness.” That trough is the uncanny valley.

    Mori’s original graph wasn’t based on experiments. It was based on his personal observations and intuitions, which is a detail that tends to get lost in the retelling. The essay was more philosophical provocation than empirical finding, and it sat in relative obscurity for decades before roboticists and CGI animators in the 2000s rediscovered it and realized they’d been stumbling into the valley independently. The 2004 Robert Zemeckis film The Polar Express—in which Tom Hanks was motion-captured into a CGI character that audiences found deeply unsettling despite technically impressive animation—became the canonical example of uncanny valley in popular culture. The technology was extraordinary. The result was a children’s movie that gave children nightmares.

    What makes the uncanny valley genuinely interesting, rather than just a fun piece of trivia to reference when a new humanoid robot demo goes viral, is that nobody fully agrees on why it happens. And the competing explanations have very different implications for whether it can be solved.

    The competing theories

    The most commonly cited explanation is perceptual mismatch—the idea that human brains are optimized over millions of years of evolution to process human faces with extraordinary precision, and when something is close to human but slightly off, the discrepancy triggers an error signal. We’re wired to detect subtle abnormalities in faces because historically, detecting disease, deception, or unfamiliarity in the people around you was a survival advantage. A robot that’s 95 percent human-looking trips the same alarm system that would fire if you encountered a person with something wrong with them—asymmetrical facial movement, delayed eye tracking, a smile that doesn’t reach the upper face. The problem isn’t that the robot looks bad. The problem is that it looks almost right, and the remaining five percent registers as pathological.

    A second explanation focuses on categorical ambiguity. Things that sit cleanly in one category—”obviously a machine” or “obviously a human”—are psychologically comfortable because the brain knows what schema to apply. Things that fall between categories—not quite machine, not quite human—create cognitive dissonance. You don’t know whether to treat it as an object or a person, and that uncertainty is inherently aversive. This explanation has roots in psychological research on disgust responses to category violations more broadly—the same mechanism that makes chimeric creatures in horror films effective.

    A third theory, supported by a 2021 study published in Computers in Human Behavior, argues that the uncanny valley is driven specifically by the perception that a robot has feelings. Researchers found that humanoid robots are unsettling in part because people unconsciously attribute the capacity for subjective experience—what philosophers call phenomenal consciousness—to things that look human. When you look at a robot face and your brain automatically assumes something is going on behind those eyes, and then another part of your brain recognizes that nothing actually is, the collision between those two assessments produces the eerie feeling. The study demonstrated that “dehumanizing” a humanoid robot—explicitly telling people it has no feelings—significantly reduced uncanny valley responses, and this effect held up in a field study with hotel guests interacting with real robots in Japan.

    This third theory is the one with the most interesting commercial implications, because it suggests the uncanny valley isn’t purely a design problem. It’s a framing problem.

    What the latest research shows

    A February 2025 study from researchers working with Nadine, a hyper-realistic humanoid robot, tested whether equipping a robot with LLM-powered conversational abilities could reduce uncanny valley effects. Eighty participants interacted with the robot and completed pre- and post-interaction surveys. The findings were promising and limited in roughly equal measure: LLM-enhanced conversations significantly reduced feelings of eeriness and increased perceptions of pleasantness and approachability. But—and this is the part that matters for anyone trying to commercialize these things—a subset of users continued to report discomfort even after extended, high-quality conversation. The uncanny valley effect shrank. It didn’t disappear.

    The regression analysis revealed something counterintuitive: conversational naturalness and interestingness predicted willingness to continue interacting, but visual human-likeness did not. In other words, once the conversation was good enough, how human the robot looked stopped being a significant factor in whether people wanted to keep talking to it. The implications of that finding, if it replicates, are substantial—it suggests that investing in better AI conversation may be a more efficient path past the uncanny valley than investing in more realistic skin texture.

    The Columbia Engineering lip-movement paper approaches the same problem from the opposite direction. Their argument is that facial expression—particularly lip synchronization during speech—is so fundamental to human communication that getting it right is non-negotiable. Nearly half of human attention during face-to-face conversation is directed at the speaker’s lips. A robot whose mouth movements don’t match its speech triggers uncanny valley responses even if everything else looks perfect, because the mismatch between what you hear and what you see is processed as a deep wrongness. Their robot learned lip movements through self-supervised learning—watching itself in a mirror and comparing its movements to human video—rather than being explicitly programmed. The result was lip synchronization natural enough that Lipson, a self-described “jaded roboticist,” reported involuntarily smiling back at his own creation.

    The design choice nobody talks about

    Here’s the thing about the uncanny valley that tends to get overlooked in favor of the more dramatic question of “can we cross it?”: most of the companies actually building commercial humanoid robots have decided not to try.

    Look at the design language of the machines that are closest to deployment. Boston Dynamics’ Atlas has no face at all—it’s a headless torso with extraordinary agility. Agility Robotics’ Digit has a face that is deliberately stylized and cartoonish—two large “eyes” on a rounded head that reads unmistakably as friendly and unmistakably as not human. Figure’s robots have a visor-like head that evokes a motorcycle helmet more than a human face. Tesla’s Optimus has gone through several iterations, and each one has moved further from human facial features toward a smooth, featureless faceplate.

    These aren’t failures of engineering ambition. They’re deliberate design decisions informed by the uncanny valley research. If your robot’s face can’t be indistinguishable from a human face—and no current robot face can—then the safest design choice is to make it clearly, obviously, comfortably non-human. Stay on the left side of the valley. Don’t attempt the crossing. The cartoon robot, the friendly geometric face, the abstract visor—these designs maintain positive emotional responses without risking the plunge into eeriness.

    The companies pursuing hyper-realistic human faces—Hanson Robotics (creators of Sophia), Hiroshi Ishiguro’s Geminoid series—tend to be research labs and publicity vehicles rather than commercial deployment operations. Sophia is famous, but Sophia isn’t picking items in a warehouse or delivering medication in a hospital. The robots that are actually being deployed at scale are the ones that look like robots, and that’s not a coincidence.

    Whether any of this matters

    The pragmatic argument—increasingly popular among robotics engineers who are tired of the uncanny valley being treated as an unsolved existential crisis—is that the entire question is overrated for most commercial applications. A warehouse robot doesn’t need a face. A manufacturing robot doesn’t need to be likable. Even in healthcare and hospitality, where robots interact directly with humans, the evidence suggests that functional competence matters more than facial realism. The Japanese hotel study found that guests’ satisfaction with robotic service was unaffected by whether the robot was framed as human-like or machine-like, as long as it performed its job.

    The counterargument—and it’s a serious one—is that the applications where the uncanny valley matters most are precisely the applications where humanoid robots would generate the most value. Eldercare. Companionship. Therapy. Education. Customer-facing roles that require trust, rapport, and emotional connection. If you want a robot that an elderly person feels comfortable having in their home, or that a child feels safe learning from, or that a patient trusts to assist with rehabilitation, the emotional response to its face is not a trivial consideration. It’s the consideration.

    The honest answer to “will we cross the uncanny valley?” is probably “yes, eventually, for robots that are sufficiently expensive and specifically designed for contexts where crossing matters.” The Columbia lip-movement research, the LLM conversation studies, and advances in silicone skin and micro-actuator facial expression systems are all pushing the right side of the valley upward. But for the next decade of commercial humanoid robotics, the dominant strategy will almost certainly be avoidance rather than crossing—designing robots that are useful, functional, and friendly without pretending to be something they aren’t. The uncanny valley is a real phenomenon with real psychological mechanisms behind it. The most practical response, for now, is to respect it rather than try to solve it.

    We cover the uncanny valley alongside the mechanical engineering, AI integration, and commercial deployment of humanoid robots across 24 lectures in our Humanoid Robots & Drones course—including why the companies spending billions on robot bodies are making very specific choices about what those bodies look like.

  • Humanoid Robots in 2026: Who’s Building Them and How Close Are They to Useful?

    Sometime in late 2021, Elon Musk stood on a stage and announced that Tesla would build a humanoid robot. Then a person in a spandex bodysuit walked out and did a little dance. This was the prototype. Five years later, Tesla has converted its Model S and Model X production lines at the Fremont factory to manufacture Optimus Gen 3, committed $20 billion in capital expenditure for 2026, and Musk has projected annual production of one million units. On the Q4 2025 earnings call, however, Musk acknowledged that the robots currently deployed in Tesla’s own factories are not doing “useful work”—they are learning and collecting data. Which is a very expensive way of saying they’re interns.

    This is the state of humanoid robotics in March 2026: more money, more companies, more capability demonstrations, and more press releases than at any point in the history of the field—and approximately zero humanoid robots doing anything that a $40,000 industrial robot arm couldn’t do better, faster, and without falling over. The technology is real. The progress is significant. The gap between the demo reel and the work site remains enormous. And the number of people who seem to understand this gap is significantly smaller than the number of people writing breathless headlines about it.

    Here’s the actual landscape, company by company.

    Tesla Optimus

    Optimus is the most visible humanoid robot project in the world, which is what happens when the world’s most famous billionaire decides it’s going to be “more significant than the car business.” The Gen 3 iteration—revealed in early 2026—features 22-degree-of-freedom hands with 50 actuators, which is a legitimate engineering achievement. The hands can pick up an egg without crushing it, which sounds trivial until you consider how many sensors, control loops, and force-feedback calculations are running in real time to make that possible. The body remains the Gen 2 platform: 173 centimeters tall, 57 kilograms, with 28 degrees of freedom and a claimed 24-hour battery life.

    Tesla’s strategy is to treat Optimus the way it treated electric vehicles—leverage vertical integration, AI infrastructure from the Full Self-Driving program, and automotive-scale manufacturing to drive the unit cost below $30,000. At volumes exceeding one million units per year, Musk claims production costs could drop below $20,000. For context, Boston Dynamics’ Atlas has never been available for commercial purchase and estimates for comparable platforms from other companies run $100,000 to $150,000.

    The skepticism is warranted though. Musk’s timeline projections across every company he runs have a well-documented relationship with reality best described as “directionally correct but temporally delusional.” Full Self-Driving was supposed to be feature-complete in 2019. The Cybertruck was announced in 2019 and began deliveries in late 2023. Robotaxi service was supposed to launch in 2020. External customer deliveries of Optimus are now targeted for late 2026, which—adjusted for Musk Time—probably means sometime in 2028. Robotics pioneer Rodney Brooks has been publicly and specifically skeptical, and his track record on predicting the gap between robotics demos and robotics products is considerably better than Musk’s track record on predicting his own timelines.

    What Tesla does have is scale. No other humanoid robotics company has access to automotive-grade manufacturing infrastructure, a battery supply chain, an AI training cluster processing real-world data from millions of vehicles, and the capital to sustain years of losses while iterating. The question isn’t whether Tesla can build an impressive robot—they already have. The question is whether they can build one that does productive work reliably, in unstructured environments, without a team of engineers babysitting it. That’s a fundamentally different problem than building a robot that looks good on stage.

    Figure AI

    Figure is the startup that’s moved fastest from announcement to deployment. Founded in 2022, backed by Microsoft, OpenAI, Jeff Bezos, and NVIDIA—a funding roster that reads like a fantasy football lineup for the tech apocalypse—Figure has its Figure 02 robots actively deployed at BMW’s Spartanburg, South Carolina manufacturing facility. Not in a lab. Not in a demo. In a factory, doing factory work, alongside human workers.

    Figure 02 stands about 5 feet 6 inches, weighs 70 kilograms, and can lift roughly 20 kilograms. The robot uses a proprietary AI system called Helix that integrates vision, language, and tactile feedback—it can look at an object, understand a verbal instruction about what to do with it, adjust its grip based on the material’s properties, and execute the task. The OpenAI partnership gives Figure access to frontier language model capabilities for natural-language task understanding, which is a meaningfully different approach than Tesla’s end-to-end neural network strategy derived from autonomous driving.

    The BMW pilot is the real differentiator. While Tesla’s robots are collecting data inside Tesla’s own factories, Figure’s robots are performing actual tasks in a customer’s factory. That’s the gap between R&D and product, and as of March 2026, Figure is further across that gap than anyone else except arguably Agility Robotics. Figure has also announced Figure 03, targeting consumer home use by late 2026, which—if it ships on time—would make it one of the first humanoid robots marketed for residential deployment.

    Boston Dynamics Atlas

    Atlas is the robot everyone pictures when they hear “humanoid robot”—the one doing backflips, navigating obstacle courses, and recovering from shoves with unsettling grace. For over a decade, Atlas was the benchmark for what was mechanically possible in bipedal locomotion. Then in April 2024, Boston Dynamics retired the hydraulic Atlas and unveiled an all-electric version designed specifically for commercial applications.

    Electric Atlas is targeting industrial deployment in 2026, starting with Hyundai—Boston Dynamics’ parent company—at its Georgia manufacturing facility. Estimated pricing is in the $140,000 to $150,000 range, which positions it as a premium platform for high-value manufacturing tasks rather than a mass-market product. The robot’s athletic capability remains unmatched—nobody else is doing the dynamic whole-body movements that Atlas demonstrates—but the question is whether that athleticism translates into useful work or is primarily an engineering flex.

    Boston Dynamics’ history suggests caution. Spot, their quadruped robot, took years to go from viral video sensation to commercial product, and its actual use cases—construction site monitoring, industrial inspection, remote sensing—are narrower than the YouTube highlight reel implied. Atlas could follow the same trajectory: deeply impressive technically, deployed in specific high-value niches, but not the general-purpose factory worker that the hype suggests.

    The Chinese contingent

    This is the part of the landscape that gets insufficient attention in Western media. China’s humanoid robotics companies—Unitree, Agibot, Leju Robotics, XPENG’s IRON program—are shipping volume. Unitree’s G1 is projected to move 10,000 to 20,000 units in 2026, which would make it the highest-volume humanoid robot producer in the world by a significant margin. China holds an estimated 85 to 90 percent of global humanoid robot shipments.

    Unitree’s strategy is the Chinese manufacturing playbook applied to robotics: aggressive pricing, rapid iteration, and a willingness to ship products that are good enough for specific tasks rather than waiting for general-purpose perfection. The G1 is priced aggressively enough that the cost-benefit calculation works for warehouse and logistics applications where you’re replacing a human worker doing repetitive material handling. It’s not elegant. It’s not doing backflips. It’s moving boxes, and it’s doing it at a price point that makes the economics work.

    At the AW 2026 conference in Seoul in March, China’s major robotics firms demonstrated live commercial-ready systems with explicit scaling roadmaps. This isn’t the R&D phase anymore. It’s the deployment phase, and it’s happening faster in Shenzhen than in Fremont.

    What “useful” actually means

    The fundamental problem with humanoid robots in 2026 isn’t locomotion, manipulation, or AI. Those are all improving rapidly. The fundamental problem is that the world wasn’t built for robots—it was built for humans, by humans, with human proportions and human capabilities assumed at every level of design, from doorknob height to shelf spacing to the assumption that whoever is operating a forklift has the judgment to not drive it into a wall.

    A humanoid form factor theoretically solves this by being the right shape to operate in human environments without requiring those environments to be redesigned. That’s the pitch: you don’t need to retrofit your factory for robots if the robot is shaped like the worker it’s replacing. In practice, the pitch breaks down because “shaped like a human” and “capable like a human” are separated by decades of unsolved problems in manipulation, balance recovery, spatial reasoning, and the kind of contextual judgment that lets a warehouse worker notice that a shelf is about to collapse before it actually does.

    The robots that are doing useful work right now—and there are some—are doing it in structured environments with limited task variety. BMW’s Figure robots are handling specific components on specific production lines. Amazon’s warehouses use mobile robots extensively, but they’re wheeled platforms, not humanoids. The humanoid form factor adds cost and complexity that is only justified if the robot needs to navigate stairs, use human tools, or operate in environments that can’t be modified. For the vast majority of current automation needs, a robot arm bolted to the floor or a wheeled platform following a painted line on the floor is cheaper, more reliable, and less likely to fall over.

    The honest forecast: by the end of 2026, there will be thousands of humanoid robots operating in controlled industrial settings worldwide—predominantly in Chinese factories and warehouses, with smaller deployments at Hyundai, BMW, and Tesla’s own facilities. They will be doing structured, repetitive tasks. They will require significant human oversight. They will not be folding your laundry, mowing your lawn, or serving you coffee. The gap between “a robot that can pick up an egg on camera” and “a robot you’d trust to unload a dishwasher without supervision” is wider than any press release suggests, and closing it is measured in years, not months.

    We cover the full history, engineering, and trajectory of humanoid robots and autonomous drones—from Boston Dynamics’ earliest prototypes to the Chinese manufacturing surge—across our Humanoid Robots & Drones course. If the gap between demo and deployment is the part you want to understand, that’s the deep dive.