Humanoid Robotics: what’s changing now?
In our humanoid robotics market deck, you will find everything you need to understand the market
SUMMARY
Humanoid robotics is moving from staged demonstrations into an execution race around deployment, autonomy, manufacturing, cost and repeatable customer value.
The clearest change is the quality of evidence. Operating hours, parts handled, production yields and customer contracts are starting to tell us more than choreographed videos ever could.
Automotive plants have become the strongest proving ground because they combine structured environments with jobs that still need hands, mobility and some adaptation. That makes them unusually good places to expose both the strengths and weaknesses of humanoids.
Autonomy is improving fast, but the economics may depend less on whether humans ever intervene than on how rarely they need to. One supervisor helping many robots occasionally is a very different business from one operator shadowing one machine.
AI and physical training data are becoming strategic assets in their own right. Large robot fleets can generate recoveries, edge cases and manipulation data that feed back into better models, so deployment scale may increasingly improve the software as well as the hardware economics.
Hands remain one of the hardest bottlenecks. Better dexterity expands the task set, but every extra joint adds cost, sensing, heat, wiring and another place for something to fail.
Manufacturing is becoming a competitive weapon. Figure is building dedicated production lines and tracking yield, while Chinese companies such as AGIBOT are already operating at much higher unit volumes across broader robot portfolios.
China currently has the strongest supply-chain and production advantage, while the United States still concentrates many of the biggest bets on general-purpose robot intelligence, frontier compute and foundation models. Those advantages may start reinforcing each other as larger fleets create more useful physical data.
Demand is becoming more credible, but orders, reservations and realized revenue still need to be kept separate. The market has moved beyond free pilots, yet many headline commitments depend on future deployment milestones.
The most important benchmark now is boring in the best possible way: how many useful hours a robot completes, how often it fails, how quickly it recovers, what each productive hour costs and whether customers order more after the first deployment.
This market map, featured in our humanoid robotics market deck, highlights top companies and startups in the humanoid robotics market
Are humanoid robots finally leaving the demo stage?
Humanoid robots are finally doing enough real work that we can judge some of them by operating hours and output rather than polished videos.
BMW gives us one of the cleanest examples. Figure 02 spent more than 1,250 hours working at BMW’s Spartanburg plant, handling more than 90,000 sheet-metal parts and contributing to production of more than 30,000 BMW X3s. BMW says the robot worked five days a week in ten-hour shifts during the program. That is already a very different kind of evidence from a robot completing a carefully prepared task onstage.
Agility Robotics has gone further on cumulative runtime. In materials filed with the SEC, the company says Digit robots have passed 65,000 operating hours across deployments involving customers such as GXO, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre. These deployments are still narrow, but tens of thousands of hours give us something increasingly useful to measure: whether robots keep working after the novelty wears off.
Production is moving too. AGIBOT says its cumulative output passed 15,000 robots in June after reaching 10,000 only a few months earlier. Its portfolio includes more than full-size humanoids, so that number should not be read as 15,000 factory workers.
These days, the more interesting questions are how long a robot works, how often humans rescue it, how many useful cycles it completes and whether customers deploy more after the first trial.
| Company | Useful evidence we can measure |
|---|---|
| Figure | 1,250+ hours and 90,000+ parts at BMW |
| Agility Robotics | 65,000+ operating hours across customer environments |
| AGIBOT | 15,000 cumulative robots produced across its portfolio |
| Figure BotQ | 350+ Figure 03 units produced by late April |
What jobs are humanoid robots actually doing in car factories today?
Humanoid robots currently earn their keep mostly by moving, sorting and handling things inside factories, with automotive plants becoming the industry’s main proving ground.
Figure’s first BMW application involved taking sheet-metal parts from containers and positioning them for welding. Digit has focused heavily on moving totes and materials. Hexagon’s AEON is being tested by BMW around battery assembly, component production and material movement.
Manufacturers already automate highly repetitive processes such as welding and painting with fixed industrial robots. Humans remain common where the work changes enough that fixed automation becomes awkward: picking different parts, feeding production equipment, sequencing components or moving material between areas.
Figure’s newer BMW work is more ambitious than its original loading task. Figure 03 has been shown sequencing components and pulling a wheeled cart while coordinating locomotion and manipulation.
BMW is testing both Figure and Hexagon. Mercedes-Benz has worked with Apptronik. Hyundai owns Boston Dynamics and plans to use Atlas inside its manufacturing operations. Tesla intends to use Optimus internally before trying to build a much larger external business around it. Chinese automakers are running similar programs with domestic robot makers.
Car factories are especially useful because floors are flat, parts arrive through defined processes, safety procedures exist and the value of each task can be measured. Yet humans still perform many jobs that require hands, mobility and occasional adaptation.
As this chart shows, and as featured in our humanoid robotics market deck, search interest in where to buy robots has been rising steadily
Are humanoid robots actually autonomous, or are humans still helping behind the scenes?
Humanoid robots are much more autonomous today than they were recently, but human assistance still plays a meaningful role when the machines hit unfamiliar situations.
The technical jump is substantial. Figure’s Helix models connect vision, language and physical action rather than relying mainly on separately programmed movements. Helix 02 extended that approach to whole-body control, allowing the same system to coordinate walking, balance and manipulation during longer tasks.
Google DeepMind is moving in a similar direction with Gemini Robotics. Its newer models combine embodied reasoning with whole-body control and fine manipulation, and Google is testing the technology with more than 100 robotics partners and trusted testers. Apptronik and Boston Dynamics are among the companies working directly with DeepMind.
Robots can now follow natural-language instructions, manipulate previously unseen objects and complete sequences that would once have required substantial task-specific programming. Real-world autonomy remains messier. Objects move, containers arrive in the wrong position, lighting changes, people step into the workspace and mechanical parts wear.
1X makes the remaining human role unusually visible. Its NEO home robot can use an Expert Mode in which a remote human helps with tasks NEO cannot yet handle on its own. Industrial deployments often describe similar functions as supervision, fleet operations or recovery.
The economics depend heavily on how often that happens. Teleoperation becomes expensive when one remote worker effectively shadows one robot. It becomes much more workable if one person oversees many robots and intervenes only occasionally.
Those interventions can also feed the learning system. A human rescue shows the robot how a difficult situation should have been handled, giving the company another training example. Apptronik has built Robot Park facilities around repeated physical tasks, while several developers combine human demonstrations, robot trajectories and simulation.
The metric worth watching now is interventions per operating hour and whether that rate keeps falling.
Is AI becoming more important than the humanoid robot itself?
AI is currently becoming the biggest source of differentiation in humanoid robotics, while hardware is increasingly being built around the needs of the model.
Figure is spending at frontier-AI scale. Its announced Nscale infrastructure agreement could eventually involve up to 100,000 NVIDIA Vera Rubin GPUs, with Figure describing an initial compute commitment of roughly $3.5 billion and potentially more than $6 billion if expanded. Those numbers look much more like an AI-lab budget than the traditional R&D budget of a robotics manufacturer.
Google and NVIDIA are also making robot intelligence less tied to one body. Gemini Robotics is being tested across different embodiments. NVIDIA’s Isaac GR00T stack combines foundation models, simulation, synthetic data, robot-learning tools and Jetson Thor computing, while its new reference humanoid puts a Unitree body, third-party dexterous hands and NVIDIA compute into one development platform.
That opens the door to a robotics ecosystem where some companies specialize in bodies, others in models and others in applications. We are not there yet because transferring a skill between machines with different joints, hands, torque limits and sensors remains difficult.
Hardware still decides whether that intelligence survives a normal working day. Figure said its previous robot taught the company painful lessons about failures in areas such as forearm electronics, which were redesigned for Figure 03.
If you want more recent data on this point, please see our latest humanoid robotics market report.
This chart, featured in our humanoid robotics market deck, illustrates yearly venture capital funding for humanoid robotics startups
Why is robot training data suddenly such a big deal?
Physical training data has become one of the hardest assets to build in humanoid robotics because the internet never created a giant dataset showing robots exactly how objects feel, move and fight back.
Large language models had an extraordinary head start: trillions of words written by people. Robots need information about friction, force, balance, failed grasps, object weight, collisions and thousands of other physical details that are poorly represented in text and video.
Figure’s Index project is designed to collect physical interaction data at large scale. The company says its system is capable of generating roughly 35 minutes of training data every second. Apptronik is running fleets inside dedicated Robot Park facilities. NVIDIA mixes real robot demonstrations with internet video, teleoperation and synthetic trajectories created in simulation.
Google DeepMind is trying to make each datapoint more valuable by transferring knowledge across different robots. If a model learns a useful manipulation pattern on one embodiment and can adapt it to another, the industry needs fewer demonstrations for every new machine.
Fleet size could therefore create a compounding advantage. A company operating thousands of robots may gather more edge cases, recoveries and physical interactions every day than a research team can deliberately reproduce in months.
Raw hours can still fool us. A robot repeating the same easy motion does not necessarily create a valuable dataset.
Are humanoid robot hands still holding the industry back?
Humanoid hands remain one of the biggest hardware headaches because reliable human-level manipulation is still much harder than basic mobility.
A robot may need to pick a flexible cable, rotate a tool, slide a component into a tight fixture and adjust its grip when something moves unexpectedly. Human hands solve those problems constantly without conscious effort.
The hardware is expensive. Cost estimates published by Morgan Stanley and discussed by JPMorgan put dexterous hands at roughly $15,000 of the bill of materials for an Optimus-like robot using a Chinese supply chain. Actuators account for roughly another $22,000. With a non-Chinese supply chain, those estimates rise sharply.
Companies are adding more capability anyway. NEO uses highly articulated hands. XPeng’s IRON has 21 degrees of freedom per hand. NVIDIA’s latest reference humanoid uses five-finger tactile hands with 22 degrees of freedom.
Every additional joint brings more motors, transmissions, sensing, wires, heat and possible failure points. That helps explain why industrial robots will sometimes use simpler grippers when they can do the same job more reliably.
This chart, featured in our humanoid robotics market deck, shows how Agility Robotics is capturing share in humanoid robotics
How fast are humanoid robot costs actually falling?
Humanoid robot costs are falling fast enough to change the business case, although the cheap headline prices we see online still describe very different kinds of machines.
Goldman Sachs previously estimated that humanoid manufacturing costs had dropped by roughly 40% in about a year, with its broad cost range moving from around $50,000-$250,000 toward $30,000-$150,000. That decline came from improving components, a larger supplier base and designs better suited to production.
Chinese manufacturers are pushing prices lower. Unitree’s smaller G1 starts in the low tens of thousands of dollars, which makes humanoid hardware accessible to far more laboratories and developers. But a research-oriented G1 should not be compared directly with an industrial Digit expected to work long shifts inside a customer facility.
The bigger change is happening inside manufacturing. Figure redesigned Figure 03 so that more components can be stamped, die-cast or injection-molded instead of CNC-machined in small quantities. By late April, BotQ had produced more than 350 Figure 03 robots and demonstrated a one-robot-per-hour assembly cycle, compared with roughly one per day earlier in the ramp.
The quality numbers are worth watching more closely than the headline capacity. Figure reported an end-of-line first-pass yield above 80%, a 99.3% first-pass yield on its battery line, more than 500 battery packs produced and over 9,000 actuators across more than ten designs.
| Cost driver | What is changing |
|---|---|
| Actuators | Higher volumes and stronger Chinese supplier base |
| Hands | Still unusually expensive and mechanically complex |
| Structural parts | Moving from CNC machining toward casting, stamping and molding |
| Electronics | Increasing standardization around high-volume compute and sensors |
| Assembly | Dedicated lines are replacing prototype-style builds |
Why is China moving so much faster in humanoid robotics, and has it already won?
China currently has the clearest manufacturing advantage in humanoid robotics, but we still do not have enough evidence to call it the overall leader in useful general-purpose autonomy.
AGIBOT is the most obvious example of the manufacturing gap. The company reached its 15,000th cumulative robot only a few years after being founded. Omdia had already estimated that AGIBOT shipped 5,168 humanoids in the previous year, giving it about 39% of the market measured by the research firm.
UBTECH is moving up the same curve. The company says orders for its Walker S2 industrial humanoid exceeded RMB 800 million and has discussed annual capacity in the thousands. XPeng is bringing automotive manufacturing experience directly into the development of IRON.
The component economics are even more revealing. Morgan Stanley estimates discussed by JPMorgan put the bill of materials for an Optimus-like humanoid at roughly $45,500 when sourced through China versus around $131,800 through a non-Chinese supply chain. It is an illustrative model rather than the audited cost of an actual robot, but a gap of that size is hard to ignore.
China also benefits from dense manufacturing clusters. Motors, reducers, batteries, electronics, sensors, housings and machining capacity can often be sourced within the same industrial regions. McKinsey has highlighted that supplier density as a major advantage for Chinese humanoid manufacturers.
Lately, China has also started pushing outward commercially. AGIBOT has launched robot-rental offerings in Europe and, more recently, several Middle Eastern markets.
Production volume can blur the question of who is actually ahead. AGIBOT’s cumulative production includes several robot categories and deployment types. A robot shipped into education, entertainment or data collection cannot automatically be counted as the equivalent of an industrial machine completing thousands of productive shifts.
The US has a different concentration of advantages. Figure is spending billions on compute and developing its own foundation models. Google DeepMind is pushing generalist physical intelligence across several robot platforms. NVIDIA is supplying much of the training, simulation and onboard computing stack. Apptronik and Boston Dynamics are combining advanced hardware with those AI ecosystems.
As seen above, China’s lower supply-chain costs could eventually reinforce its AI position too. Large fleets create physical experience, and physical experience can improve models.
Today, China leads industrialization and production volume, while some of the strongest bets on general-purpose robot intelligence remain concentrated in the United States.
If you want more recent data on this point, please see our latest humanoid robotics market report.
This chart, featured in our humanoid robotics market deck, illustrates yearly funding for humanoid robotics startups
Has Tesla fallen behind in humanoid robots?
Tesla has fallen behind the leading humanoid companies on publicly demonstrated deployment, even though Optimus still has one of the strongest paths to massive scale if the robot works.
Competitors have produced clearer operational evidence. Agility has logged tens of thousands of customer hours. Boston Dynamics has moved the electric Atlas into production and has initial deployments tied to Hyundai and Google DeepMind. Chinese manufacturers are already building humanoids and related robots in much larger quantities.
The company has been converting space at Fremont from Model S and Model X production toward Optimus manufacturing and has described its next robot generation as the version meant for serious production. Early units are expected to feed Tesla’s own development and data-collection efforts, including its internal Optimus Academy.
What changed recently is the tone around timing. Tesla stopped repeating some of the more aggressive near-term volume language that previously accompanied Optimus.
Tesla still has unusually strong manufacturing advantages. The company already knows how to build motors, batteries, electronics and complicated electromechanical systems at very high volume, and it controls factories where Optimus can be deployed internally.
But today, Optimus deserves to be judged on what Tesla has actually deployed. By that standard, Tesla is chasing several competitors rather than leading them.
Are companies finally paying real money for humanoid robots?
Companies are finally committing serious money to humanoid robots, although much of that demand is still contracted future business rather than revenue already earned.
Agility provides some of the best evidence because its SEC disclosures force more detail than we usually get from private robot companies. Agility says it has more than $300 million of multi-year Digit v5 orders, including a contract covering 1,000 robots under three-year robot-as-a-service terms, subject to deployment milestones.
UBTECH says Walker S2 orders have passed RMB 800 million. Those commitments suggest industrial customers are willing to budget real capital for humanoids rather than simply host experiments.
Consumer robotics is also producing early demand. 1X priced NEO at $20,000 for early-access ownership and has separately discussed a $499 monthly subscription. The company said the first 10,000 units of annual production capacity were reserved within five days of opening orders.
We should not turn those numbers into revenue prematurely. An order subject to milestones can be delayed, and a reservation can disappear.
Still, the quality of demand has improved sharply.
This chart, featured in our humanoid robotics market deck, compares the main business model options for humanoid robot manufacturers
Can humanoid robots already beat human workers on cost?
Humanoid robots can already make financial sense for a few high-utilization jobs, but reliability decides the calculation far more than the robot’s purchase price.
Agility’s public materials give us a useful example. Its illustrative Digit economics use roughly $200,000 for the robot, around $20,000 for deployment and about $36,000 per year for software and maintenance. Across five years, that works out to roughly $400,000 of direct spending before financing and site-specific costs.
A $400,000 machine sounds expensive until it works across multiple shifts. Digit has been designed around long daily operating schedules supported by battery swaps and charging. If a robot performs several thousand productive hours each year, the hardware cost gets spread across a very large amount of work.
Human labor costs also extend beyond hourly pay. Employers carry payroll costs, benefits, training, hiring, overtime, supervision and turnover. A machine that can cover work currently spread across two or three shifts has a much easier ROI calculation than one replacing a person for six hours a day.
Reliability can destroy that model quickly. If the robot regularly stops, needs an engineer nearby or completes the task much more slowly than a worker, the apparent savings disappear.
If you want more recent data on this point, please see our latest humanoid robotics market report.
Why are humanoid robots still painfully slow at simple tasks?
Humanoid robots are still slow at many everyday tasks because moving faster makes almost every physical failure mode harder at the same time.
Picking up a component sounds trivial to us. A robot has to identify the object, estimate its pose, choose where to grab it, move without hitting anything, make contact, apply the right force, confirm the grasp and keep control of the object while its whole body moves.
The industry could make many demos look faster by accepting more failures. Factories usually want the opposite.
Sanctuary AI gave a useful example when it reported more than 99.5% success on an automotive wire-plugging task while reaching a 2.54-second cycle time benchmarked against the production process. Combining cycle time with success rate tells us much more than watching a robot move quickly in one edited clip.
One failure every 20 cycles may look fine during a two-minute demonstration and become a disaster after thousands of repetitions.
This chart, featured in our humanoid robotics market deck, shows the revenue mix across customer segments in the humanoid robotics market
Do humanoid robots really need legs?
Many useful humanoid robots probably do not need legs, and companies are becoming much less religious about copying the human body exactly.
BMW’s work with Hexagon makes this obvious. AEON combines a human-like upper body with a wheeled base. On a smooth factory floor, wheels consume less energy and avoid the constant balance problem that comes with walking on two feet.
For some environments, legs remain extremely valuable. Digit is designed to move through spaces built for workers. Figure and Atlas are targeting wider sets of tasks where steps, obstacles and changing floor layouts may eventually matter.
The right body depends on the job. A robot working beside one production machine may need no mobility whatsoever. A warehouse robot may be better on wheels. A machine expected to move through stairs, doorways and clutter has a much stronger case for legs.
Sanctuary AI has taken the idea further by putting its Physical AI software onto existing industrial robot hardware rather than insisting every deployment use Phoenix.
If you want more recent data on this point, please see our latest humanoid robotics market report.
Are humanoid robots moving into homes faster than expected?
Home humanoid robots are arriving earlier than expected as products people can order, while the technology needed for a genuinely independent household robot is still far behind.
1X has pushed furthest into that gap with NEO. The robot is deliberately light, soft-covered and quiet, with hardware designed around close interaction with people rather than industrial payloads. Its early-access ownership price is $20,000.
The interesting part is the service model. NEO can perform some tasks autonomously while difficult jobs can be handled through remote Expert Mode. Early customers are therefore buying a robot whose capabilities can include human assistance behind the scenes.
Figure is also moving aggressively toward homes. Its newer hardware emphasizes softer materials, improved battery safety, wireless charging and better sensing, while many recent Helix demonstrations take place around kitchens and domestic objects.
Homes are an unforgiving benchmark. Factory engineers can control lighting, floor conditions, containers and workflow. A home has pets, children, clothes on the floor, fragile objects, unfamiliar packaging and people changing where things belong.
We still have no evidence that a home robot can autonomously provide the broad, dependable help people imagine when they hear “robot butler.”
This chart, featured in our humanoid robotics market deck, shows how factory humanoid robot technology has evolved over time
Is safety becoming a bigger problem as humanoid robots improve?
Humanoid robot safety is becoming harder precisely because the machines are getting useful enough to work near people instead of remaining behind cages.
BMW’s experience shows what happens when a humanoid leaves the lab. Its Figure deployment produced lessons around revised safety concepts, including additional barriers and partitions, along with changes to connectivity inside the plant.
The challenge grows as humanoids become more general. Traditional industrial robots usually repeat defined motions in controlled cells. A mobile AI-controlled machine may walk around people, change tasks and respond to situations that were not programmed one by one.
Standards are moving in that direction. Updated ISO 10218 requirements have strengthened the framework around industrial robots and collaborative applications, but humanoids combine several categories that were historically treated separately.
Figure also had to work through battery-safety questions for Figure 03. The company says no existing battery standard perfectly matched the application, so it worked with an OSHA-recognized testing laboratory on an appropriate UL testing approach covering mechanical, electrical and thermal abuse.
If a general-purpose robot needs a custom fenced area every time it enters a factory, much of its flexibility disappears.
Is humanoid robotics becoming a real business or another investment bubble?
Humanoid robotics is becoming a real business while parts of its valuation boom are clearly pricing in years of success that have not happened yet.
Capital has moved astonishingly fast. Figure raised more than $1 billion at a $39 billion post-money valuation. Apptronik’s Series A expanded beyond $900 million as investors backed its Apollo platform and physical-AI strategy. Agility’s planned public-market transaction values the company in the billions before the sector has established anything close to automotive-scale revenue.
Those valuations are difficult to justify from present humanoid sales alone.
At the same time, dismissing the whole category as hype has become much harder. There are paid deployments, multi-year robot contracts, dedicated production lines, falling component costs, foundation models built specifically for robotics and customers moving beyond one-off lab tests.
The speculative part is the speed of the extrapolation. Investors are effectively assuming that today’s narrow factory tasks lead to much broader work, that autonomy keeps improving, hardware failure rates fall, manufacturing costs decline and enough customers deploy fleets large enough to create software-like recurring revenue.
Humanoid robotics has already become commercially real. What remains speculative is how large the profit pool becomes and how many of today’s highly valued robot companies actually capture it.
If you want more recent data on this point, please see our latest humanoid robotics market report.
In our humanoid robotics market deck, we identify pain points entrepreneurs should prioritize
What is really changing in humanoid robotics right now?
Humanoid robotics is moving from a capability race into an execution race, and that is the biggest change happening right now.
A few years ago, walking reliably, lifting boxes or manipulating unfamiliar objects could define a leading humanoid. Those capabilities are spreading. The harder questions today are whether a company can manufacture hundreds or thousands of robots, keep them working for long shifts, lower intervention rates and get customers to expand deployments.
AI is accelerating that transition. Helix, Gemini Robotics and GR00T show that robot behavior is becoming less dependent on manually programming each task. Larger physical datasets are giving models more ways to learn. Cross-embodiment research suggests some of that intelligence may eventually move between different robot bodies.
Manufacturing is advancing just as quickly. Chinese supply chains are driving down component costs, AGIBOT has reached five-digit cumulative production and Figure has already moved Figure 03 onto a dedicated manufacturing system with measurable yield improvements. As pointed out above, those numbers do not prove equivalent commercial productivity.
The shape of the robot is changing too. Wheeled humanoids, specialized grippers and physical-AI systems running on conventional industrial machines are all competing with the classic two-legged, five-fingered design.
The leading humanoid company will need reliable hardware, cheap manufacturing, enough deployments to generate huge amounts of useful physical data and AI that steadily reduces the amount of human help required.
Today, several companies can demonstrate parts of that formula. Nobody has put all of it together at massive scale yet.
OUR METHODOLOGY
We broke humanoid robotics into the dimensions that actually shape whether the industry is progressing: real-world deployment, autonomy, AI and training data, hardware capability, manufacturing, cost, commercial demand, geographic advantages, safety and business maturity.
We gave more weight to evidence tied to execution than to polished demonstrations. Sustained operating hours, parts handled, production output, manufacturing yields, task success rates, customer commitments and disclosed economics were treated as stronger evidence than a single successful video or an announced production target.
We also kept different forms of progress separate before comparing companies. Manufacturing scale tells us whether robots can be built repeatedly. Deployment tells us whether they survive real environments. Autonomy tells us how much work happens without human recovery. Orders, reservations and realized business were treated as different levels of commercial evidence rather than blended together.
Where comparisons could become misleading, we compared like with like. Research-oriented hardware was not treated as economically equivalent to an industrial robot expected to work long shifts, and broad portfolio production was not automatically counted as full-size humanoid deployment. Geographic leadership was assessed across supply chains, manufacturing, AI capability and real-world use rather than reduced to one headline ranking.
We prioritized first-hand and checkable material wherever possible: customer deployment data, SEC filings, company technical releases, manufacturing updates, official investor disclosures and international safety standards. Institutional research was used selectively when it added a useful cost or market framework that individual company disclosures could not provide on their own.
Key sources include: BMW Group on Figure 02’s Spartanburg deployment and AEON testing, Agility Robotics’ SEC filing on operating hours and contracted Digit v5 orders, Figure on Helix 02 and full-body autonomy, Figure on BotQ production, yields and Figure 03 output, Figure on Index and large-scale physical-data collection, Figure on its Nscale compute agreement, Google DeepMind on Gemini Robotics, NVIDIA on the Isaac GR00T reference humanoid, 1X on NEO autonomy and Expert Mode, AGIBOT on its 15,000th robot production milestone, Boston Dynamics on Atlas moving into production, Apptronik on its $935 million Series A and Apollo scale-up, Sanctuary AI on automotive wire-plugging performance, ISO 10218-1:2025, and ISO 10218-2:2025.
This chart, featured in our humanoid robotics market deck, shows the regional revenue mix across Europe, Asia, North America, Africa, and South America in the humanoid robotics market
Related blog posts
- Humanoid Robotics: what are the biggest challenges now?
- Humanoid Robotics: what is getting real adoption now?
- Humanoid Robotics: what is actually working now?
- The latest funding news in the humanoid robotics market
- What are the top startups in the humanoid robotics market?
