Signals Inbox·July 16, 2026·Humanoid Robotics

Humanoids: what are the latest milestones?

Humanoid robotics has moved beyond isolated demonstration videos. The latest milestones now include measurable factory work, commercial deployments, real production ramps and systems designed to keep learning after deployment.

Humanoid Milestone Tracker

We reviewed 30 meaningful humanoid robotics milestones announced during the past 18 months.

The tracker focuses on four questions: did the robot gain a materially harder capability, enter a real operating environment, generate measurable work, or become easier to manufacture and deploy at scale?

We classify milestones by the gap they close. Field-changing establishes an unusually strong new benchmark for the sector. Major removes an important obstacle to deployment or scale. Meaningful extends a company’s capabilities or commercial position in a credible way. Incremental improves an existing system without greatly changing its readiness. Mostly signaling generates attention but adds little operational evidence.

Humanoid robotics milestones over the past 18 months

Q1Which humanoid company has progressed the fastest recently?

Figure has progressed the fastest recently because no other humanoid company in our 18-month dataset connected as many capability, deployment and manufacturing milestones into one continuous sequence.

Figure accounts for 14 of the 30 tracked milestones, compared with four for 1X and three each for Agility and AgiBot. Raw volume can exaggerate progress when a company publishes frequently, so we also separated the events by importance and type. Figure produced four field-changing milestones, four major milestones and six meaningful ones. More importantly, its milestones cover every stage from model development to customer work.

The sequence began with the original Helix generalist controller. Figure then expanded into logistics, laundry and dishwasher handling before introducing Figure 03 and unified full-body control through Helix 02. During the same period, Figure 02 accumulated 1,250 hours at BMW, Figure 03 entered the same customer environment, and BotQ built more than 350 units while increasing production cadence 24-fold.

Agility has stronger evidence than Figure on one narrow commercial metric: Digit moved more than 100,000 totes at GXO before adding Toyota as a second customer. However, Agility generated only three milestones in the tracker and disclosed no comparable robot-generation change or production acceleration. Boston Dynamics entered the period later with a product-version Atlas and named customers, while Apptronik only recently introduced Apollo 2 and its Robot Park fleet-learning system.

Based on both the number and continuity of milestones, Figure currently has the strongest acceleration. It represents 47% of the complete dataset and 57% of all field-changing milestones, while covering more stages of the deployment chain than any competitor.

Q2What has changed most in humanoid robotics recently?

The clearest recent change is that humanoid progress has moved from isolated capabilities toward complete deployment systems.

Ten of the 30 milestones in our dataset concern a new physical or AI capability. Another 15 concern production, customer deployments, commercial availability, fleet learning, batteries or development platforms. The balance has moved beyond teaching a prototype one more visible trick.

Three connected developments show the shift particularly well. Figure followed its Helix capability work with a new robot generation, 350 completed units and another BMW deployment. Agility converted its original GXO deployment into 100,000 tote movements and then added Toyota. UBTECH paired mass production of Walker S2 with autonomous battery swapping, addressing both robot availability and manufacturing volume.

The same pattern appears in data collection. Figure created Project Go-Big and added Brookfield environments. Apptronik opened Robot Park around fleets of Apollo 2 robots. AgiBot’s SOP framework allows deployed robots to keep learning from real interactions. These companies are building repeatable feedback loops rather than treating every demonstration as a separate achievement.

Only one-third of the milestones are primarily capability demonstrations. Half now concern the machinery surrounding deployment: factories, customer access, fleet data and operating continuity.

Q3Is China or the United States leading humanoid robotics today?

The United States leads the public milestone record today, while China shows greater breadth than our verified dataset can fully measure.

US companies account for 23 of the 30 milestones, compared with six from China and one from Germany. The US advantage becomes even larger at the top of our scale: all six field-changing milestones came from Figure or Agility. These include 1,250 BMW operating hours, 100,000 totes moved at GXO, Figure’s 24-fold production-rate increase and the two generations of Helix.

China’s six milestones are concentrated around industrialization. AgiBot reported its 1,000th robot, continuous industrial operation and a deployment-learning framework. UBTECH introduced autonomous battery swapping and moved Walker S2 into mass delivery. Unitree packaged H2 Plus around NVIDIA’s complete humanoid-development stack. Five of the six Chinese milestones therefore relate to production, deployment infrastructure or easier development rather than isolated demonstration videos.

There is one annoying but important problem with the comparison: disclosure. Figure and Agility publish unusually specific runtime, throughput and production numbers. Chinese companies often announce fleet size, deliveries or factory entry without releasing the same task-level operating data. Our tracker penalizes milestones that cannot be measured independently or compared consistently, so the 23-to-six result is not a complete count of everything happening in both countries.

Based on verifiable public evidence, the United States currently leads in measured customer work and integrated AI-to-deployment progress. China looks closer in manufacturing depth and supplier breadth than the headline count suggests, but it needs more comparable operating disclosures to overturn the US lead.

Q4Which humanoid company turns technical progress into deployments most effectively?

Figure currently converts technical progress into deployments most effectively, while Agility achieves the highest commercial conversion from a much narrower technical program.

We identified a deployment conversion when a capability or product milestone was followed by manufacturing output, customer arrival, a commercial agreement or measurable work. Figure produced eight capability and product milestones, then connected them to four deployment or manufacturing outcomes: Figure 02’s BMW results, Figure 03 production, Figure 03’s return to BMW and the Brookfield deployment pipeline.

That gives Figure roughly one downstream deployment or production milestone for every two technical milestones. The chronology is clear. Figure 02 generated factory experience, Figure 03 absorbed the design lessons, BotQ built hundreds of units, and the new generation returned to BMW with Helix 02.

Agility’s ratio is even tighter because the company publishes fewer technical demonstrations. One broad capability expansion sits alongside two strong commercial results: 100,000 totes at GXO and a Toyota agreement. Agility is very efficient at converting a focused product into customer activity, although its wider capability and manufacturing progression remains less extensive than Figure’s.

1X currently sits at the other end. NEO Gamma, consumer orders, a production facility and a new hand form a coherent product sequence, but we still lack cumulative household operating hours, completed tasks or delivered-unit figures. Apptronik has customer relationships, funding, Robot Park and Apollo 2, yet it has not released comparable throughput from customer sites.

Figure wins on total conversion depth. Agility deserves the narrower distinction of extracting the most commercial proof from the fewest visible technical milestones.

Q5Which humanoid companies are compounding progress rather than producing isolated milestones?

Figure, Agility and 1X currently show the clearest compounding milestone chains, although each chain compounds a different part of the market.

Figure has the longest chain in the tracker. Helix improved generalist manipulation, BMW supplied real production data, Figure 03 incorporated the lessons, Helix 02 unified whole-body control, BotQ accelerated output, and the new robot returned to BMW. We can trace at least six directly connected milestones rather than a collection of unrelated demonstrations.

Agility’s chain is shorter but commercially cleaner. Digit entered GXO, exceeded 100,000 tote movements, expanded its handling range and converted a Toyota pilot into a commercial agreement. Four milestones reinforce the same proposition: a focused logistics robot can be extended gradually across workflows and customers.

1X is compounding toward the home. NEO Gamma established the safer domestic design, NEO became orderable, the California factory entered production, and the new 25-degree-of-freedom hand expanded the object range. The missing link is household performance data from paying users. Until that appears, the chain demonstrates product readiness more clearly than autonomous usefulness.

AgiBot and UBTECH also show connected industrial progress, but gaps remain in their public evidence. AgiBot links production, industrial operation and continuous learning without detailed throughput. UBTECH connects battery autonomy and mass delivery without publishing long-duration customer results.

Figure has produced the strongest compounding pattern because its hardware, AI, production and customer milestones keep increasing the value of earlier milestones.

Q6Which humanoid company has the biggest gap between visibility and deployment evidence today?

1X currently has the largest gap between public visibility and measurable deployment evidence among companies with a credible commercial product.

The company generated four milestones in our tracker: NEO Gamma, consumer ordering, the NEO Factory and its new dexterous hand. Together, they cover product design, sales, manufacturing and hardware capability. What remains absent is the metric connecting them: delivered robots completing repeated household work.

The contrast with Figure and Agility is pretty stark. Figure published runtime, parts handled, vehicles supported, units manufactured and production cadence. Agility published tote throughput and named a second commercial customer. 1X currently publishes no equivalent total for autonomous chore completions, household operating hours, intervention frequency or robots working inside customer homes.

NEO’s use of remote experts makes these missing figures particularly important. Human supervision can create immediate usefulness while making a demonstration appear more autonomous than the underlying system. The ratio between autonomous completion, remote assistance and outright failure would tell us far more than another household video.

Boston Dynamics and Apptronik also attract considerable attention without much public throughput data, but their products are earlier in the new deployment cycle. 1X has already accepted consumer deposits and started full-scale production, which raises the evidence standard.

This gap does not mean NEO lacks value. It means 1X has progressed further through the commercial funnel than its publicly measurable autonomy has progressed through the evidence funnel.

Q7Are humanoid milestones becoming more important or simply more frequent?

Humanoid milestones are becoming both more frequent and more consequential, with the strongest improvement coming from manufacturing and measured deployment.

The first six months of our 18-month window produced nine milestones. Only two were field-changing: the original Helix release and AgiBot’s 1,000th robot. Much of the activity involved new platforms, planned factories or expanded capabilities.

The most recent six months produced 12 milestones, including three field-changing events: Helix 02, more than 350 Figure 03 units and the 24-fold production-rate increase embedded in that ramp. The period also added Toyota as a Digit customer, product-version Atlas shipments, Apptronik’s Robot Park and full-scale NEO production.

Frequency increased by roughly 33%, from nine milestones in the earliest six-month period to 12 in the latest one. The composition improved at the same time. Five of the recent 12 events involved customer deployment, actual manufacturing or a fleet-data facility, compared with two of the earliest nine.

There is still plenty of milestone inflation. A new manipulation video and an operating-hour disclosure cannot carry the same weight. The stronger conclusion comes from the rising number of milestones that expose production rates, customer names, completed work or physical facilities.

Q8Which humanoid company has the most balanced progress today?

Figure currently has the most balanced humanoid program, while Agility remains the most deployment-heavy and 1X the most consumer-focused.

We assessed balance across four dimensions: capability, deployment, manufacturing and commercial reach. Figure recorded milestones in all four. Its tracker entries include six capability advances, two measured BMW deployments, two manufacturing milestones and several commercial or data-access relationships.

Agility scores extremely well in deployment and commercial proof but remains thinner in manufacturing disclosures and broad capability progress. Its three milestones revolve around expanded material handling, 100,000 totes and Toyota. That concentration is commercially credible, though less balanced than Figure’s program.

1X covers capability, product design, consumer ordering and manufacturing. It therefore looks balanced on paper, but deployment evidence remains the weak side. UBTECH covers capability and production well, while public customer-performance data remains limited. AgiBot spans manufacturing, deployment and learning infrastructure but provides fewer comparable operational metrics.

Boston Dynamics and Apptronik may become more balanced quickly, although most of their latest milestones concern preparation for deployments rather than completed work.

Figure is the only company in our dataset with at least two meaningful milestones in every major dimension. That breadth explains why it leads several comparisons without owning every individual benchmark. Agility still holds the cleaner commercial-throughput proof, and 1X has advanced furthest toward an orderable household product.