Signals Inbox·July 28, 2026·Humanoid Robotics

Figure vs. Apptronik: who is ahead now?

Figure is ahead of Apptronik today because it has manufactured more robots, documented more customer work and put its current generation into the field; Apptronik has the partners and capital to close the gap, but Apollo 3 still has to turn that setup into operating fleets.

Q1Why do Figure and Apptronik keep getting compared?

Figure and Apptronik keep getting compared because both want to sell the same first product: a general-purpose humanoid worker for factories and warehouses.

The overlap is unusually direct. Figure has worked with BMW on automotive production and is entering retail logistics through Catalyst Brands. Apptronik has placed Apollo robots with Mercedes-Benz, Jabil and GXO for manufacturing and warehouse trials. Both companies initially target work such as moving components, preparing assembly kits, sorting packages, loading machines and delivering materials to production lines.

Their long-term ambitions are similar too. Figure wants Figure 03 to move from commercial facilities into homes. Apptronik sees Apollo expanding from factories and warehouses into retail, healthcare and domestic work. Both are betting that one robot design can eventually learn many jobs instead of being engineered around a single workflow.

The comparison has become more relevant lately because the two companies have reached very different milestones. Figure 03 has returned to BMW for a new logistics task, and Figure has signed another commercial deployment agreement. Apptronik has opened its nearly 90,000-square-foot Robot Park, introduced Apollo 2 publicly and raised its total Series A financing above $935 million. Yet Apptronik still describes Apollo 2 mainly as the platform preparing its future Apollo 3 commercial fleet.

So the useful question is no longer which robot looks better. It is which company has moved further from training robots to putting them to work.

Q2Why is it hard to say whether Figure or Apptronik is winning?

Figure currently leads on documented robot work, while Apptronik has the broader group of industrial and technology partners.

Those advantages do not fit neatly on the same scoreboard. Figure has published production quantities, factory operating hours, components handled and manufacturing yields. Apptronik has announced major relationships with Google DeepMind, Mercedes-Benz, Jabil and GXO, but it has released far fewer numbers showing what Apollo robots accomplish at those customer sites.

The products are also at different stages. Figure presents Figure 03 as its current scalable robot. Apptronik says Apollo 2 is gathering data, training AI systems and supporting customer pilots before Apollo 3 becomes its main commercial product. A feature-by-feature comparison would put a commercial generation against a bridge to the next one.

Neither company discloses the figures that would settle the argument quickly. We do not know their humanoid revenue, paying fleet size, recurring customer spend, contract backlog, intervention rate, cost per working hour or gross margin per robot.

For now, the best evidence is what the robots already do: how long they work, how autonomous they are, whether customers expand deployments and whether the companies can manufacture reliable units repeatedly.

Q3Has Figure done more real factory work than Apptronik?

Figure has done far more documented factory work than Apptronik so far.

During Figure 02’s deployment at BMW’s Spartanburg plant, the robots accumulated more than 1,250 operating hours, loaded over 90,000 sheet-metal parts and contributed to the production of more than 30,000 BMW X3 vehicles. Figure says the robots operated during ten-hour weekday shifts on an active assembly line.

Across the reported runtime, that works out to roughly 72 loaded parts per hour, or one part every 50 seconds. The calculation does not tell us the speed of each individual robot because Figure never disclosed the fleet size. It does show sustained production work rather than a short demonstration prepared for cameras.

Figure’s original task was fairly narrow. The robots picked known sheet-metal pieces and placed them into welding fixtures. BMW and Figure still had to meet a strict factory rhythm, position parts within a five-millimeter tolerance and avoid repeatedly stopping the line.

The latest Figure 03 assignment is harder. The robot must sort components that may arrive shifted, rotated or partly hidden, place them into the correct trolley positions and pull a heavy cart while repositioning its feet. We still need operating results from this project, but BMW has clearly moved Figure from fixed loading toward a more variable logistics workflow.

Apptronik has active relationships with Mercedes-Benz, Jabil and GXO, yet none has produced comparable public numbers. We do not know how many Apollo robots operate at these sites, how many hours they have worked or how often they complete tasks without intervention.

Public factory evidence disclosed by Figure and Apptronik

Q4Does Apptronik have better customers than Figure?

Apptronik has the broader industrial network, but Figure has gone further with its best customer.

Apptronik’s customer and partner list is difficult to dismiss. Mercedes-Benz provides a demanding automotive environment and has invested in the company. GXO operates large logistics facilities where repetitive material handling is a natural first market for humanoids. Jabil can both test Apollo inside its factories and manufacture robots for Apptronik. Google DeepMind contributes one of the world’s strongest robotics AI teams.

Each relationship covers a different part of the commercialization problem. Mercedes and GXO offer real workplaces, Jabil offers production expertise and Google offers frontier AI models. Apptronik does not have to build every capability alone.

The weakness is conversion. Mercedes originally described the project as a pilot. GXO initially evaluated Apollo in a laboratory before any distribution-center deployment. Apptronik now says Apollo 2 systems operate at customer and partner Robot Parks, although it has not published enough usage data to show how far those programs have progressed.

Figure has fewer major customer names, but its BMW relationship has already moved across two robot generations. BMW used Figure 02 in production and has now brought in Figure 03 for a more difficult workflow. Figure has also signed Catalyst Brands for distribution and logistics work at a Nevada facility serving brands such as JCPenney, Aéropostale and Brooks Brothers.

Brookfield gives Figure another route into customers. The asset manager is an investor in Figure and Catalyst Brands, owns or manages large residential, office and logistics portfolios and can introduce Figure to operating businesses within that network.

Apptronik has opened more doors. Figure has walked further through the most important one.

Q5Is Figure 03 more ready for customers than Apptronik Apollo 2?

Figure 03 is more commercially ready today than Apptronik Apollo 2.

Figure designed Figure 03 around its Helix AI system, larger-scale manufacturing and work in commercial facilities. The same generation is now being produced in the hundreds, tested internally and assigned to external customer projects.

Apptronik describes Apollo 2 differently. Its latest Robot Park announcement calls Apollo 2 a data-collection and training platform that has been working behind the scenes for more than a year. The robot comes in bipedal and wheeled configurations and is being used at Apptronik facilities, customer sites and Google DeepMind.

The lessons and data from Apollo 2 are feeding Apollo 3. Apptronik describes that upcoming generation as its commercial fleet, and chief executive Jeff Cardenas declined to tell Business Insider when it would be ready.

Apollo 2 is still a serious engineering platform. It has upgraded motors, batteries and sensors, and customers can use it to test real workflows. But Apptronik’s own language makes the stage gap plain: Figure is deploying the generation it wants to scale, while Apptronik is using its current generation to finish the one it wants to scale.

Q6Is Figure manufacturing humanoid robots faster than Apptronik?

Figure is manufacturing humanoid robots at a completely different disclosed scale from Apptronik.

In its latest production update, Figure said BotQ had delivered more than 350 Figure 03 robots. The assembly line improved from one completed robot per day to a one-hour production cycle in less than 120 days.

That hourly cycle should not be mistaken for thousands of customer shipments. Figure allocates robots across research, data collection, household training and commercial development, and it has not disclosed how many units are earning revenue. A production line can also reach a one-hour cycle without maintaining that rate continuously through every shift.

The supporting manufacturing figures are still unusually detailed for this industry. Figure reported an end-of-line first-pass yield above 80%, a 99.3% battery-line yield, more than 500 completed battery packs and over 9,000 actuators across more than ten designs. Each robot undergoes more than 80 verification tests before approval.

An 80% final first-pass yield leaves plenty of room for improvement. Roughly one robot in five still needs some correction before passing the first time. The important bit is that Figure can now measure defects across hundreds of complete machines and adjust the process.

Apptronik’s Jabil agreement gives it a credible way to manufacture at scale without building the entire system alone. Jabil has global factories, supply-chain experience and established quality controls. Apptronik has not disclosed the number of Apollo units built by Jabil, the current assembly rate or the expected initial capacity for Apollo 3.

Latest disclosed humanoid manufacturing measures

Q7Does Figure have better robot AI than Apptronik?

Figure currently has the stronger autonomous humanoid system, while Apptronik has the stronger outside AI partner.

Figure’s Helix 02 controls the robot’s hands, arms, torso, balance and movement through one learned system. Figure has shown the robot tidying rooms, loading a dishwasher, folding laundry and manipulating packages. These are company-produced demonstrations, so they cannot replace independent customer results. Still, the range shows Figure testing one control architecture across very different forms of manipulation.

The standout is whole-body coordination. Figure 03 can move its feet and torso while using both hands instead of treating walking and object handling as separate routines. That is essential for work where the robot cannot remain planted in front of a fixed table.

Apptronik has a major advantage through Google DeepMind. Gemini Robotics can adapt across several types of robot, including Apollo, ALOHA platforms and Franka arms. Google says one model can learn from several embodiments, which may let useful improvements travel between different machines.

The public Apollo evidence is earlier. Business Insider reported that much of the activity inside Robot Park still involves remote operators guiding or closely supervising the robots. Apptronik uses those sessions to create training data, which makes sense, but it has not shown an Apollo fleet completing long autonomous shifts at customer speed.

For now, Figure has turned robot AI into more visible autonomous behavior. Apptronik has access to a formidable model ecosystem; the commercial benefit has not arrived yet.

Q8Is Figure or Apptronik building the stronger data flywheel?

Figure currently owns the stronger data flywheel because its robot, factory, training system and AI model sit inside the same company.

Every Figure 03 produced at BotQ can be assigned to research, data collection or customer work and feed what it learns back into Helix. Figure can then change the hands, cameras, actuators or electronics when the data exposes a physical weakness. No negotiation across separate hardware, model and manufacturing companies is needed.

The Brookfield partnership adds human training data. Brookfield manages more than 100,000 residential units and large office and logistics portfolios. Figure has begun collecting first-person human video in those environments so Helix can learn movements and tasks without every example being demonstrated directly on a robot.

Apptronik’s approach is broader. Robot Park collects data through teleoperation, simulation and autonomous attempts. Apollo 2 units at partner and customer locations can add examples from factories and warehouses, while Google DeepMind can train across several robot designs.

That breadth could become a major advantage. The trade-off is control. Gemini Robotics is a shared model platform rather than an Apollo-only system, so Google can apply related advances to other robots too.

Figure’s loop is narrower and more expensive to build. It is also more proprietary, easier to coordinate and already connected to a larger disclosed fleet.

Q9Is Apptronik Apollo easier to deploy than Figure 03?

Apptronik Apollo is easier to fit into a conventional warehouse or factory than Figure 03.

Apollo 2 can use either legs or a wheeled base. For many early warehouse tasks, wheels are the practical choice. They consume less energy, remove the risk of a heavy robot falling and fit more easily within existing industrial mobile-robot safety practices.

A customer that only needs a robot to move between conveyors, carts and workstations gains little from paying the technical cost of humanlike walking. Apptronik can begin with wheels, gather data and add bipedal mobility where the environment genuinely requires it.

Apollo also uses four-hour replaceable batteries. Apptronik says battery swapping can keep a robot available across a workday of up to 22 hours, assuming batteries and operators are ready. The company has also designed physical safety zones that pause the robot when people or objects enter defined areas.

Figure 03 is lighter and softer than its predecessor, with protective coverings, tactile fingers and several layers of battery safety. It remains fully bipedal, which lets it step, reposition its body and operate in spaces built around human movement. That flexibility brings more balance, power and safety problems with it.

Apptronik has the better architecture for cautious first deployments. Figure has the more ambitious one for places where wheels eventually become a limitation.

Q10Which company is closer to solving humanoid robotics’ real bottleneck?

Figure is closer to solving the bottleneck that decides near-term adoption: keeping an autonomous robot useful for long periods without constant human help.

Humanoid robotics already has enough impressive videos. Customers need predictable cycle times, few interventions, safe operation and repair costs low enough to produce a financial return. A robot that completes ten different tasks once can be less useful than a robot that completes one task 10,000 times reliably.

Figure has reached the stage where field use and larger production batches expose ordinary engineering problems. It has discussed component failures, supplier quality, manufacturing yield, burn-in testing and design changes based on operational experience. Those details are not glamorous. They are much closer to what factories pay for.

Apptronik is attacking the data shortage. Robot Park exists because humanoid AI lacks the giant training datasets available to language models. The facility runs throughout the week and gives operators a controlled place to generate examples for many tasks.

That work is necessary, but more training data cannot solve every mechanical or economic problem. Apptronik still has to show that Apollo can maintain performance through customer shifts, avoid frequent resets and create enough value to justify its purchase and support costs.

Neither company publishes mean time between failures, intervention frequency or cost per productive hour. Figure gets the lead because it has moved further into the reliability problem, though the hardest financial evidence is still missing from both.

Q11Is Figure moving faster than Apptronik right now?

Figure has moved faster than Apptronik lately, especially along the path from research to a manufactured robot and then into customer work.

Figure has introduced a new robot generation, increased its manufacturing rate, expanded Helix into whole-body household tasks and put the same hardware and AI stack into new commercial workflows. The progress crosses hardware, software, production and deployment instead of stopping in the research lab.

The company also retires old hardware quickly. Figure ended the wider Figure 02 program after Figure 03 became available and carried the lessons into the new design. That is expensive, but it prevents engineering teams from spending years supporting several immature generations.

Apptronik has accelerated too. Its recent financing gives it nearly $1 billion in total capital, Robot Park expands the supply of training data and Apollo 2 creates a common platform for wheeled and legged tests. Those are strong foundations for Apollo 3.

The difference is where the momentum ends. Figure’s latest development cycle ends with its newest robot entering customer environments. Apptronik’s ends with more capital, more training infrastructure and a better platform for developing the robot that comes next.

Q12What can Figure do that Apptronik cannot easily copy?

Figure’s tightly integrated technology stack is harder for Apptronik to copy than Apptronik’s partner-based approach is for Figure to recreate.

Figure controls the humanoid hardware, hands, actuators, battery, manufacturing line, fleet systems, training pipeline and Helix models. Its engineers can change a camera position because the AI struggles with an obstructed view, redesign electronics after a reliability problem or alter production testing when failures appear in the fleet.

Apptronik controls important parts of its own system, including Apollo’s hardware, actuator technology, Artemis controls and Fleet Connect software. It combines those products with manufacturing help from Jabil and AI models from Google DeepMind.

The modular strategy may let Apptronik adopt outside improvements more quickly and avoid some of Figure’s capital spending. Large customers may also prefer a system that can work with several technology providers rather than one closed stack.

Apptronik’s Google relationship offers less exclusivity than it once appeared to. Google DeepMind now works with several robot platforms, and Boston Dynamics has also announced a Gemini Robotics partnership for Atlas. Apptronik still receives valuable access and expertise, but Google’s progress will not automatically become an Apollo-only advantage.

Figure carries more execution risk because it has chosen to build almost everything. If the system works, it owns more of the technology, data and manufacturing knowledge that competitors would need to reproduce.

Q13Which company can afford the humanoid race longer?

Figure has more financial firepower, while Apptronik has a much easier valuation to justify.

Figure raised $675 million in its Series B and later announced more than $1 billion in Series C commitments at a $39 billion post-money valuation. Those two rounds alone amount to at least $1.675 billion.

Apptronik raised $415 million in its initial Series A and another $520 million in an extension, bringing the round above $935 million and its total funding close to $1 billion. Business Insider reported that the latest financing valued Apptronik above $5.5 billion.

Figure has therefore raised roughly 1.7 times as much across those major rounds, while its valuation is around seven times higher. Investors are giving Figure far more credit for becoming the category leader than its financing advantage alone would suggest.

The extra money is useful because Figure has chosen the expensive route. It operates its own manufacturing facility, builds robot fleets before mass customer revenue, develops proprietary AI models and funds data collection across several environments.

Apptronik can share part of that burden with Jabil and Google DeepMind. Its lower valuation also means Apollo 3 can create substantial investor value without immediately overtaking Figure.

Neither company discloses enough revenue to calculate a credible valuation multiple. Figure can spend more aggressively, but it has much less room for delays, weak customer demand or disappointing robot economics.

Q14Are Figure and Apptronik leading the whole humanoid market?

Figure is ahead of Apptronik today, but Agility Robotics currently has the strongest public record for commercial humanoid deployment.

Agility recently disclosed that its Digit robots had accumulated more than 65,000 operating hours across commitments involving nine customer facilities. Digit has moved more than 100,000 totes at one GXO location, and Agility says it has secured over $300 million in multi-year Digit v5 orders, subject to contractual milestones.

Those figures go beyond what either Figure or Apptronik currently reports. Figure has shown more dexterous and general behavior than Digit, but Agility has stronger evidence that customers are committing money and operating robots repeatedly in logistics environments.

Boston Dynamics is becoming a more serious commercial competitor too. The company has started manufacturing the product version of Atlas, says all of its initial deployments are committed and plans fleets for Hyundai and Google DeepMind before adding more customers. Hyundai’s manufacturing network gives Atlas a direct route into large-scale industrial testing.

Figure’s claim to leadership is therefore specific. It appears furthest ahead among companies trying to combine dexterous hands, whole-body learned control, in-house manufacturing and broad general-purpose ambitions.

Apptronik remains one of the strongest challengers, particularly because of Mercedes, Jabil and Google. Its deployment record does not yet put it at the front of the wider market.

Q15Who is ahead now: Figure or Apptronik?

Figure is clearly ahead of Apptronik right now.

The three decisive gaps are commercial readiness, manufacturing output and documented autonomous work. Figure is producing its current robot generation in meaningful quantities, using it in customer environments and publishing far more evidence about what its machines accomplish.

Apptronik has built an excellent position around Apollo. Its wheeled platform is practical, its partner network is broad and its funding is large enough to keep it in the race. Google DeepMind, Mercedes-Benz and Jabil could help Apptronik close gaps in AI, customer access and production faster than a normal robotics startup could.

The problem is that much of Apptronik’s case still points forward. Apollo 2 is gathering the data for Apollo 3. Robot Park is training machines for customer work. Jabil is preparing a manufacturing route. Major customers are testing use cases. Figure’s strongest evidence describes work and hardware that already exist.

Apptronik could change the answer by launching Apollo 3, disclosing fleet numbers and showing robots completing long customer shifts with little supervision. A repeat order or multi-site expansion from Mercedes, GXO or Jabil would carry much more weight than another partnership announcement.

Figure now has to prove that its lead can become a business. It needs productive use from its Figure 03 fleet, results from the new BMW workflow, sustained operations under the Catalyst agreement and steady quality as BotQ produces more robots.

Another customer record with substantial autonomous runtime would make Figure’s lead difficult to close. A fast Apollo 3 rollout across several existing partner sites, with comparable operating results, could still pull Apptronik back into the race within a few development cycles.

Figure vs. Apptronik: current leader by criterion