In January 2026, Figure 02 worked five days a week for ten hours per shift at BMW's Spartanburg plant, helping produce more than 30,000 BMW X3s over ten months. That's not a demo. That's a humanoid robot doing repetitive factory work at automotive production scale. The race to deploy humanoid robots in 2026 has moved past the prototype stage, and the gap between what companies claim and what they've actually shipped is now measurable in real production hours, not YouTube views.
| Company/Robot | Real Deployments (April 2026) | Production Status | Est. Price |
|---|---|---|---|
| Agility Digit | 7+ units at Toyota Canada (RAV4 logistics); moved 100,000+ totes at GXO | Commercial revenue-generating | $250,000 |
| Figure 02 | 30,000+ vehicles supported at BMW Spartanburg; single-digit units at BMW | 1 robot/hour at BotQ factory | $100k-$250k (estimated) |
| Boston Dynamics Atlas | 2026 units fully committed to Hyundai RMAC and Google DeepMind | Production began Jan 2026 | ~$150,000 (estimated) |
| Tesla Optimus | Musk confirmed Jan 28, 2026: units "not doing useful work," remain in "R&D phase" | Fremont production prep starts Q2 2026 | Target: under $20,000 |
| Apptronik Apollo | Operating in Mercedes-Benz and Jabil facilities (pilot scale) | $935M raised, production ramp underway | Not disclosed |
| Unitree G1/H1 | ~5,500 humanoids shipped in 2025; targeting 10k-20k in 2026 | Volume leader by units shipped | Started below $20,000 (base spec) |
Agility's Digit is the only humanoid robot generating revenue from productive commercial work as of April 2026. That sentence matters because it cuts through the noise. Figure robots supported production of 30,000+ vehicles at BMW Spartanburg, logging 1,250+ hours and moving 90,000+ parts, but those were pilot deployments focused on proving viability, not revenue-generating contracts. Agility signed commercial agreements with Toyota Motor Manufacturing Canada for seven Digit units working RAV4 logistics, while Digit moved over 100,000 totes at GXO warehouses under paying contracts.
Boston Dynamics took a different path. The company unveiled the production Atlas at CES on January 5, 2026, and began production immediately at its Boston headquarters, with a 30,000-unit/year factory planned for 2028. But all 2026 Atlas deployments are fully committed to Hyundai's RMAC facility and Google DeepMind, with no third-party customers until 2027. This isn't a weakness; it's a controlled ramp strategy. Hyundai Motor Group plans to deploy tens of thousands of robots across its factories and is investing $26 billion in U.S. operations.
Tesla's position is harder to parse. On January 28, 2026, Musk stated that Optimus units are "not doing useful work" and remain in "R&D phase," confirming the credibility gap between demos and deployable autonomy. Yet Tesla plans to start preparations for the first large-scale Optimus factory in Q2 2026 in Fremont, designed for a capacity of one million robots per year, replacing the Model S and Model X production lines. The timeline disconnect is stark.
The term "general-purpose humanoid" gets thrown around constantly. Every humanoid robot company claims to be building a general-purpose robot, but every actual deployment is hyper-specialized: totes for Digit, parts kits for Apollo, battery cells for Optimus, car parts for Figure 02. Here's what they're proven to handle in real environments:
Material Handling and Logistics: This is the only category with multi-site, revenue-generating deployments. Digit operates for up to 8 hours on a single charge, enough for a full warehouse shift, handling tote movement and bin picking. Figure 02 retrieved and positioned sheet metal parts for welding, operating five days a week for ten hours per shift. BMW required Figure to load sheet metal parts within 37 seconds and complete the entire task in 84 seconds, though Figure AI didn't disclose whether those targets were met.
Automotive Assembly Support: The most documented industrial deployments as of 2026 are Figure AI's pilot at BMW (material handling and parts transfer) and Agility's Digit at Amazon fulfillment centers, covering a narrow range of material handling tasks, not high-speed, high-precision assembly. Boston Dynamics says Atlas will perform high-precision sequencing at scale in 2028 and complex assembly tasks slated for 2030, pushing real assembly work years into the future.
Battery and Autonomy: Atlas features 56 degrees of freedom, fully rotational joints, a 2.3-meter reach, and can lift up to 50 kg (110 lbs), and the robot autonomously swaps its own batteries for continuous operation without downtime. Apollo uses hot-swappable battery packs, each with a four-hour runtime, allowing it to continue working with a simple battery change instead of a plug-in charge. Tesla hasn't disclosed Optimus battery specs publicly.
The Optimus story in 2026 is a study in contradictions. On one hand, Tesla is designing a first-generation line for a capacity of one million robots per year in Fremont, while at Gigafactory Texas, a second-generation line targets annual production capacity of 10 million robots. On the other, Tesla targeted 5,000 Optimus units in 2025 and delivered only hundreds, a >90% shortfall reported by The Information and confirmed by Musk on Q2 2025 earnings call.
Based on Tesla's track record of 12-24-month timeline slippage, meaningful internal deployment is realistic by late 2026 to early 2027, contingent on AI software bridging the gap from data collection to reliable autonomous operation. The company is collecting training data by recording human workers at its Fremont and Austin factories, using video-based learning rather than explicit programming. Tesla Optimus is not available for third-party purchase as of April 2026, and scale deployment of 10,000+ working Optimus units is realistic in the 2028-2029 window, not 2026 or even 2027.
The pricing target remains aggressive. Elon Musk has stated the robot could eventually be priced at less than $20,000, dramatically lower than competitors. But total cost of ownership, including integration engineering, environment preparation, and ongoing maintenance for a platform with limited published MTTF data, will significantly exceed the robot unit price.
Figure 02's BMW deployment offers the clearest window into what a humanoid pilot actually looks like at automotive scale. Within 6 months of bringing up Figure 02, the company delivered robots to the plant and began testing; within 10 months, full deployment launched on an active assembly line, running every single working day. Across more than 1,250 operational hours, Figure 02 recorded minimal hardware failures while generating critical data that informed Figure 03.
The failure data is instructive. The robot's forearm was the top hardware failure point at BMW. For Figure 03, the company completely re-architected the wrist electronics to eliminate both the distribution board and dynamic cabling, with each wrist's motor controller now communicating directly with the main computer. This is the kind of iteration that only happens with real production hours.
The two partners are now evaluating where else the next-generation Figure 03 could create additional value in production, and BMW confirmed on February 27, 2026, that it is deploying humanoid robots at its plant in Leipzig, Germany, marking the first time Physical AI of this kind has entered a European automotive production environment. The Leipzig deployment uses AEON, a robot from Hexagon Robotics, not Figure, showing BMW is platform-agnostic.
The Boston Dynamics Atlas is the world's first enterprise-grade humanoid robot, and as of March 2026, it's officially in production. The current production Atlas (2026) is fully electric, using custom high-powered actuators supplied by Hyundai Mobis; the switch from hydraulic to electric actuation dramatically improved reliability, efficiency, and industrial viability. This wasn't an incremental update. It was a complete platform redesign aimed at manufacturability and cost.
Hyundai is not just a parent company; it is the first and largest customer, announcing that Atlas will be deployed at the Hyundai Motor Group Metaplant America (HMGMA) in Georgia by 2028 for sequencing tasks. The newly announced Robot Metaplant Application Center (RMAC), described as a "data factory," is designed to build the world's most complete dataset for training humanoid manufacturing skills and will serve as the engine for deploying tens of thousands of robots across HMG facilities worldwide.
The AI partnership with Google DeepMind is equally significant. Google DeepMind's deployment will focus on training Atlas with foundation models, Large Behavior Models for general-purpose humanoid intelligence, cutting-edge research that could unlock capabilities far beyond current industrial tasks. This positions Atlas as both a commercial product and a research platform for embodied AI.
Apptronik raised massive capital in 2026. The company raised a $520 million Series A-X funding round, bringing its total Series A to more than $935 million and total capital raised to nearly $1 billion. The funding round valued Apptronik at $5 billion, and the company has inked partnerships with Mercedes-Benz, GXO Logistics, and Jabil, plus a strategic partnership with Google DeepMind to build the next generation of humanoid robots powered by Gemini Robotics.
Apollo is already operating in Mercedes-Benz and Jabil facilities, though these remain pilot-scale deployments. Apollo's torso can mount on bipedal legs, a wheeled base, or a stationary platform using the same core robot, a modularity that hedges against the unsolved question of which mobility approach wins for different factory tasks.
Unitree takes a completely different approach: volume over premium pricing. Unitree shipped ~5,500 humanoids in 2025, the current global leader by volume. The company is targeting 10k-20k units in 2026. Unitree's G1 started below USD 20,000 in early 2025 at base specifications, dramatically undercutting Western competitors but focusing on research and hobbyist markets rather than industrial deployments with paying contracts.
The humanoid robot industry in 2026 has crossed a threshold that matters more than any individual product spec: multiple companies are now logging real production hours in paying customer environments. Humanoid robots are performing real work on factory floors in 2026, but only at a handful of pilot sites, for a narrow set of tasks, at cycle times and reliability levels that traditional industrial robots cleared a decade ago.
Three changes are driving real deployments. First, sim-to-real training has matured. Large-scale simulation-to-real training pipelines now allow robots to accumulate billions of virtual manipulation experiences before touching a physical object. Second, the cost structure is finally approaching viability. Industry analyst estimates place Figure AI and Agility Robotics platforms in the USD 100,000-250,000 range per unit including integration support, expensive but no longer prohibitive for high-value tasks. Third, labor shortages are forcing adoption timelines forward in logistics and manufacturing.
But the honest assessment hasn't changed much from previous years. Humanoids may supplement cobots in specific material handling and inspection roles over the next several years, not replace them. None yet operates at automotive-line production rates. The companies winning in 2026 are the ones with real customers paying real money for robots doing boring, repetitive tasks reliably. Everything else is still R&D with a good demo reel.
| Robot/Company | Key Specs | Proven Capabilities | AI/Compute Platform | 2026 Status |
|---|---|---|---|---|
| Agility Digit | 8-hour battery; 50 lb payload (next-gen); non-traditional humanoid form | Revenue-generating work; 100,000+ totes moved; 7 units at Toyota RAV4 plant | Agility Arc platform (fleet management) | ISO functional safety certification targeted mid-late 2026 |
| Figure 02/03 | ~70 kg; ~170 cm; 20 kg payload; 41 DOF | 30,000+ vehicles supported at BMW; 1,250+ hours logged | OpenAI collaboration / Helix AI | BotQ factory: 1 Figure 03 per hour |
| Boston Dynamics Atlas | 56 DOF; 50 kg payload; 2.3m reach; self-swapping batteries | Material handling, order fulfillment; autonomous navigation | Google DeepMind models | Production began; 2026 units committed to Hyundai/DeepMind |
| Tesla Optimus | 5'8" tall; 125 lbs; 20 lb carry / 150 lb deadlift; 22 DOF hands (Gen 3) | Demonstrated autonomous navigation and battery cell sorting at Fremont/Giga Texas | Cortex 2.0: 250 MW (April 2026); vertically integrated actuators/chips/sensors | "Not doing useful work" (Musk, Jan 28, 2026) |
| Apptronik Apollo | 71 DOF; 25 kg payload; hot-swap 4-hour batteries | Gross manipulation, box/crate/tote moving; 25 kg payload handles majority of warehouse items | NVIDIA GR00T platform | Pilots at Mercedes-Benz and Jabil |
| Unitree G1/H1 | Base price <$20k; focus on locomotion | High-speed locomotion; runs, jumps, recovers from pushes in lab/demo environments | Intel RealSense cameras | ~5,500 shipped in 2025; volume leader |
Agility's Digit is the only humanoid robot generating revenue from productive commercial work as of April 2026, with paying contracts at GXO, Toyota, and Amazon. By sheer unit count shipped globally, Unitree leads with ~5,500 humanoids shipped in 2025, though most target research and hobbyist markets.
No. On January 28, 2026, Elon Musk stated that Optimus units are "not doing useful work" and remain in "R&D phase". Tesla plans to start preparations for the first large-scale Optimus factory in Q2 2026 in Fremont, but production at scale remains months to years away.
Unitree's G1 started below $20,000 at base specifications, while Figure AI and Agility Robotics platforms are estimated at $100,000-250,000 per unit including integration support. Boston Dynamics Atlas is estimated at approximately $150,000 per unit. Tesla targets under $20,000 eventually but hasn't shipped commercially.
As of 2026, documented industrial deployments cover material handling and parts transfer, a narrow range of tasks, not high-speed, high-precision assembly processes. Every actual deployment is hyper-specialized: totes for Digit, parts kits for Apollo, battery cells for Optimus, car parts for Figure 02.
Unitree is targeting 10k-20k units in 2026, making it the volume leader. Tesla's Fremont line is designed for one million robots per year, but based on Tesla's 12-24-month timeline slippage track record, meaningful production volumes are realistic by late 2026 to early 2027.
Atlas is designed to tolerate water exposure, operate in temperatures from -20°C to 40°C, and includes safety features such as human detection that allow it to work without physical barriers. Some humanoid robot suppliers including Agility are involved in developing ISO 25785-1, which would provide more clarity and permissions for the unique attributes of humanoids over general ISO industrial robot standards like ISO 10218.
1X expects to deliver the first NEO home robots to US early-access customers in 2026, with subscription units to follow. Tesla has discussed home use but hasn't announced availability. Most 2026 deployments remain focused on industrial and logistics environments, not households.
Agility's Digit is the only humanoid robot generating revenue from productive commercial work as of April 2026, with paying contracts at GXO, Toyota, and Amazon. By sheer unit count shipped globally, Unitree leads with ~5,500 humanoids shipped in 2025, though most target research and hobbyist markets.
No. On January 28, 2026, Elon Musk stated that Optimus units are "not doing useful work" and remain in "R&D phase." Tesla plans to start preparations for the first large-scale Optimus factory in Q2 2026 in Fremont, but production at scale remains months to years away.
Unitree's G1 started below $20,000 at base specifications, while Figure AI and Agility Robotics platforms are estimated at $100,000-250,000 per unit including integration support. Boston Dynamics Atlas is estimated at approximately $150,000 per unit. Tesla targets under $20,000 eventually but hasn't shipped commercially.
As of 2026, documented industrial deployments cover material handling and parts transfer, a narrow range of tasks, not high-speed, high-precision assembly processes. Every actual deployment is hyper-specialized: totes for Digit, parts kits for Apollo, battery cells for Optimus, car parts for Figure 02.
Unitree is targeting 10k-20k units in 2026, making it the volume leader. Tesla's Fremont line is designed for one million robots per year, but based on Tesla's 12-24-month timeline slippage track record, meaningful production volumes are realistic by late 2026 to early 2027.
1X expects to deliver the first NEO home robots to US early-access customers in 2026, with subscription units to follow. Tesla has discussed home use but hasn't announced availability. Most 2026 deployments remain focused on industrial and logistics environments, not households.