Humanoid Robotics Leaderboard Tracker (2026)
In our humanoid robotics market deck, you will find everything you need to understand the market
Summary
The humanoid robotics race in April 2026 does not have a single winner. It has several, each leading on a different part of what winning actually means today.
Three leaders stand out. Agility Robotics' Digit leads on commercial and operational dimensions. Figure's 02 and 03 lead on AI capability and autonomy. Unitree and AgiBot lead on shipment volume and cost. No robot leads on more than four of the eleven dimensions we looked at.
Digit carries the most commercial weight. More than 100,000 totes moved at GXO under a paid Robots-as-a-Service contract is the single cleanest proof point in the category. Signed contracts with Toyota Canada and Mercado Libre in February 2026 add multi-customer depth no rival has matched.
Figure wins on AI and autonomy. Helix 02's unedited 4-minute, 61-action dishwasher sequence is the longest public humanoid end-to-end run. Figure 02's 1,250 documented hours at BMW Spartanburg is the most specific reliability record any humanoid vendor has published.
Tesla Optimus is conspicuously absent from almost every leader row despite the highest valuation and capex. Musk himself confirmed on the Q4 2025 earnings call that Optimus is not yet doing productive work. That single admission resets the conversation.
The short version is that humanoid robotics is fragmenting by segment, not consolidating around one winner. Logistics, factory, and consumer will likely crown different specialists within 24 months.
This market map, featured in our humanoid robotics market deck, highlights top companies and startups in the humanoid robotics market
Humanoid robotics leaderboard tracker
This tracker compares the major humanoid robots across 11 evidence-based questions. Each question asks which robot has the clearest public proof of a specific capability today, not which one has the most ambitious roadmap.
The table below names a winner and a runner-up for each question, with the specific evidence that earned each position. For broader market context and multi-year forecasts, we go deeper in our humanoid robotics market deck.
| Question | Winner | Why is the winner | Runner-up | Why the runner-up |
|---|---|---|---|---|
| Clearest proof it can do useful work today | Digit (Agility Robotics) | Digit crossed 100,000 totes moved in live paid operation at GXO's Spanx warehouse. It added signed commercial RaaS deals with Toyota Canada and Mercado Libre in February 2026. No other humanoid has dated productive revenue across multiple named customers. | Figure 02 (Figure AI) | Figure 02 logged 1,250 hours at BMW Spartanburg and contributed to 30,000 X3 vehicles. But the deployment ended in late 2025 and Figure 03 has not yet replicated it at multiple paying customers. |
| Clearest proof it can learn new tasks quickly today | Figure 03 (Figure AI) | Helix and Helix 02 showed new objects and tasks acquired via natural-language prompt with shared neural weights. The 61-action dishwasher demo was unedited and end-to-end, far beyond scripted routines. No rival has published a comparable dated artifact. | Digit (Agility Robotics) | Digit uses NVIDIA Isaac Sim to acquire new workflows in hours, and has actually shifted between putwall and tote-recycling skills at customer sites. But its public artifacts are less visually dramatic than Helix 02's room-scale demo. |
| Clearest proof it can finish tasks autonomously today | Figure 03 (Figure AI) | Helix 02's unedited 4-minute, 61-action dishwasher sequence is the longest public humanoid end-to-end run. Figure released it without resets, teleoperation, or prompting between steps. This is the strongest single-session autonomy evidence in the dataset. | Digit (Agility Robotics) | Digit's 100,000+ totes at GXO imply large cumulative autonomous cycles without a teleoperator in the loop. But Agility has not published per-run success rates or full unedited multi-hour footage that matches Figure's headline demo. |
| Clearest proof it can operate safely around people today | Digit (Agility Robotics) | Digit passed an OSHA-recognized NRTL field inspection at a live e-commerce fulfillment site in November 2025. That is the only regulator-recognized safety validation on any humanoid today. ISO functional safety certification is targeted mid-to-late 2026. | Apollo (Apptronik) | Apollo ships with force-control architecture and a configurable Perimeter Zone, and runs alongside Mercedes, GXO, and Jabil workers. But it has no externally certified safety credential, so its safety case rests mainly on vendor design claims. |
| Clearest proof it can operate with low failure rates today | Figure 02 (Figure AI) | Figure 02 ran 10-hour weekday shifts at BMW with three tracked goals: cycle time, placement accuracy, and interventions. 1,250 hours and 90,000 parts over eleven months is the most specific failure-adjacent dataset published. No humanoid has disclosed MTBF or error taxonomies. | Digit (Agility Robotics) | Digit's 100,000-tote GXO record and reported ~98% success rate in Amazon testing suggest strong reliability. But Agility has not published denominators, intervention counts, or a formal error taxonomy, so the conclusion remains indirect. |
| Clearest proof it can keep working over thousands of hours today | Figure 02 (Figure AI) | Figure 02 accumulated 1,250 operating hours on one BMW deployment over eleven months. That is the only humanoid with a dated, publicly disclosed per-deployment hours counter. The data point is specific, customer-confirmed, and multi-month. | Digit (Agility Robotics) | Digit has run full-time at GXO for over a year across a growing fleet, likely exceeding Figure 02 in cumulative hours. But Agility publishes tote counts rather than explicit hours, so the endurance claim is implied rather than stated. |
| Clearest proof it can generalize across different tasks today | Figure 03 (Figure AI) | Helix 02 uses a single unified neural network for locomotion, manipulation, and balance across varied tasks. Figure has shown pills, syringes, bottle caps, dishwashers, laundry, and cluttered-bin picks. Breadth under one stack is wider than any rival's. | Apollo (Apptronik) | Apollo handles inspection, sorting, kitting, lineside delivery, fixture placement, and sub-assembly at Jabil. That is meaningful task breadth in manufacturing. But it is narrower than Figure's cross-domain Helix demos and still inside one industrial context. |
| Clearest proof it can generalize across different environments today | Digit (Agility Robotics) | Digit operates across GXO, Spanx, Amazon, Schaeffler, Toyota Canada, and Mercado Libre sites. That spans apparel logistics, automotive manufacturing, and e-commerce fulfillment across multiple countries. No rival has this breadth of dated multi-customer operation. | Apollo (Apptronik) | Apollo runs in Mercedes-Benz Hungary, GXO, and Jabil environments, covering automotive and electronics manufacturing. But customer count and geographic spread are smaller than Digit's, and scaled revenue-generating operation is not yet documented. |
| Clearest proof it can replace labor cost-effectively today | Digit (Agility Robotics) | Digit is the only humanoid with multiple signed paid multi-year contracts implying customer-internal ROI approval. Third-party reporting cites $10 to $12 per hour versus $30 per hour human labor at Amazon. But no humanoid has published audited unit economics. | None clearly proven | Every vendor relies on future-price or RaaS narratives rather than disclosed cost per productive hour. Current warehouse humanoids replace only 0.3 to 0.4 FTE, implying marginal ROI, with 0.8 to 1.0 FTE replacement not expected until 2027 to 2029. |
| Clearest proof it can be deployed with minimal integration effort today | Digit (Agility Robotics) | Agility Arc integrates Digit with existing WMS and AMR systems, and the same stack has been redeployed across six named customer sites. Repeatability across sites is the strongest cross-customer evidence of manageable integration work. Vendor-reported hours-not-weeks adaptation adds to this. | Apollo (Apptronik) | Apollo has been stood up at Mercedes, GXO, and Jabil using configurable mobility bases and swappable batteries. But every public deployment still required multi-month pilot phases, and Apollo lacks Digit's cross-customer repetition count. |
| Clearest proof it can scale without huge costs today | G1 / A2 (Unitree / AgiBot) | Unitree shipped ~5,500 units and AgiBot 5,168 in 2025 at prices as low as $13,500. AgiBot hit 10,000 cumulative units by March 2026. No Western vendor has matched this delivered-volume-at-low-cost record. | Digit (Agility Robotics) | Agility's RoboFab factory is designed for up to 10,000 Digit units per year and has started customer deliveries. But volumes remain far below Chinese leaders, and the ~$250,000 unit price keeps total manufacturing scale modest by comparison. |
This chart, featured in our humanoid robotics market deck, compares the main business model options for humanoid robot manufacturers
Insights
We built this leaderboard by aggregating tier-one evidence across every humanoid currently in commercial production. That means direct customer disclosures, regulator filings, named vendor statements with dates and sites, and reporting traceable to primary sources, across eleven capability dimensions. From pulling that dataset together, clear patterns emerged. The insights below are the most useful ones we found.
- The humanoid race is not one race but two partially decoupled races. Deployment maturity and capability frontier are led by different robots. Any single-ranking comparison conceals this and produces the wrong conclusion.
- No humanoid leads on more than four of eleven dimensions. This points to segment-based consolidation, not a single winner. Logistics, factory, and consumer will likely crown different specialists within 24 months.
- The sharpest signal that a humanoid claim is weak is the absence of a denominator. Hours, tote counts, success rates, and parts moved are the currency of credible progress. Demo videos without denominators should be weighted 5 to 10 times lower.
- Customer-originated press releases carry 2 to 3 times more weight than vendor-originated ones. BMW, GXO, Toyota, and Mercado Libre disclose counts, hours, and dates. Vendors emphasize valuations, demos, and future targets.
- Capital intensity has decoupled from deployment evidence. Tesla has the highest humanoid capex and the lowest productive-work record. Valuation is not a leading indicator of commercial progress in this sector.
- The most meaningful 2026 demos are long-horizon, not new-task. Figure's 61-action dishwasher run and Digit's 100,000-tote count matter because they stack cycles. Single-task can-do-X clips no longer move the credibility needle.
- Teleoperation disclosure is a credibility signal that runs opposite to marketing instinct. Vendors who disclose human-in-the-loop frequency are more trustworthy than those who claim pure autonomy without the data. Honest teleop numbers are the precondition for believing any autonomy-improvement claim.
- Safety evidence bifurcates along one rule. If a humanoid operates only in segregated zones, vendor safety claims are not yet falsifiable at scale. No 2026 evidence supports a human-safe-at-scale conclusion for any humanoid.
- Shipment leadership is nearly orthogonal to commercial-work leadership. The world's most-shipped humanoids are research and consumer units with 2-hour batteries and 2-kg payloads. Units shipped is not a useful proxy for units doing enterprise-grade work.
- Signed multi-year commercial contracts are a stronger signal than piloted deployments. They encode customer-internal ROI approval even when dollar figures are undisclosed. Count of paying multi-customer contracts beats TAM forecasts as an early success metric.
- Every real deployment followed an 8-to-12-month three-phase arc of lab, proof-of-technology, and live pilot. Any humanoid available now that has not completed this arc is at least a year from productive deployment. Plug-and-play integration narratives should be discounted heavily.
- Vendor timelines slip by approximately 2x on average. Applying that factor to current 2027 commercial-scale claims yields 2028 to 2029 effective delivery. This should be the working baseline for planning, not vendor press releases.
- Regulatory firsts compress into durable moats that capability firsts do not. OSHA and ISO validations embed a year of auditor and customer work that later entrants must redo from scratch. The 2026 to 2027 leader will likely be whichever vendor accumulates the most certified operating hours.
- Robots-as-a-Service pricing is doing quiet strategic work across the dataset. It transfers upfront capital risk to the vendor and keeps list prices opaque. This is why no humanoid has disclosed true cost-per-hour economics, and it defers rather than answers the cost-effectiveness question.
- Current warehouse humanoids replace 0.3 to 0.4 FTE per unit, not 1.0. Any business case built on 1:1 labor substitution today is mathematically premature. One-FTE replacement is a 2027 to 2029 story, not a 2026 one.
- Industrial arms deliver 95 to 99 percent uptime, and no humanoid has published a comparable number. Roughly 10x the joint count implies multiplicative failure exposure. Crossing the 99 percent reliability threshold will need a hardware generation change, not just better AI.
- All home-robot claims rely on vendor-controlled footage. Zero independent third-party household reviews exist for any humanoid as of April 2026. This is the weakest form of evidence and should be discounted heavily.
- The strongest durable judgment rule is a four-part filter. A humanoid claim is credible only if it is dated within 90 days, tied to a named paying customer, includes a quantitative denominator, and survives absence of disclosed teleoperation. Only Digit at GXO passes all four cleanly today.
This chart, featured in our humanoid robotics market deck, shows how Agility Robotics is capturing share in humanoid robotics
Which humanoid has the clearest proof it can do useful work today?
Today, Agility Robotics' Digit shows the strongest evidence of doing useful work. No other humanoid has dated productive revenue across multiple named paying customers.
Digit crossed 100,000 totes moved in live commercial operation at GXO's Spanx warehouse under a paid Robots-as-a-Service contract. Agility announced this milestone in November 2025. The milestone is customer-confirmed and dated, which is rare in the category.
Agility added seven-unit deployments at Toyota Motor Manufacturing Canada and Mercado Libre's San Antonio facility in February 2026. Schaeffler and Amazon are also paying customers. That multi-customer breadth is the decisive factor.
Figure 02 runs a close second on work done, but the BMW Spartanburg deployment ended in November 2025. Apollo is in active pilots at Mercedes, GXO, and Jabil but has not disclosed commercial unit throughput. Optimus, Atlas, and NEO have no productive external revenue. Musk admitted on the Q4 2025 earnings call that Optimus is primarily for learning, not productive tasks.
By the way, we break down which humanoid platforms have the strongest commercial proof across automotive, logistics, and retail in our humanoid robotics market deck.
Which humanoid has the clearest proof it can learn new tasks quickly today?
Today, Figure 03 shows the strongest evidence of learning new tasks quickly. Its Helix 02 system acquires new objects and skills via natural-language prompt on a single set of neural-network weights.
Figure announced Helix in February 2025 and Helix 02 in January 2026. The 4-minute, 61-action dishwasher demo was unedited and end-to-end. No rival has published a comparable dated artifact.
Digit runs second. Agility reports using NVIDIA Isaac Sim to acquire new workflows in hours rather than weeks and has shifted between putwall and tote-recycling skills at customer sites. But the public artifacts are less dramatic than Figure's. Boston Dynamics announced fleet-level skill transfer for Atlas at CES 2026, but without external customer evidence.
Unproven elements for Figure include whether Helix 02 capabilities generalize to new homes, hold across units, and persist without re-demonstration. No third party has replicated the demo. The evidence is medium, not strong.
Which humanoid has the clearest proof it can finish tasks autonomously today?
Today, Figure 03 owns the strongest single-session autonomy evidence, and Digit owns the strongest cumulative one. The two lead in different ways, Figure on one unedited demo and Digit on repeated real operation.
Helix 02's 4-minute, 61-action run is the strongest single-session artifact. Figure released it as unedited with no intervention. For repeated autonomous operation inside a real workflow, Digit is ahead. More than 100,000 totes moved at GXO implies large cumulative autonomous cycles without a continuous teleoperator.
Optimus has no public autonomous-task proof, as Musk himself acknowledged in Q4 2025. Atlas had its first public product-version demo in January 2026 and has no customer-site autonomy data. 1X NEO explicitly ships with human-teleoperator Expert Mode fallback.
No humanoid has published full unedited multi-hour runs, exception-handling telemetry, and per-run success rates on the same task across thousands of trials. That is the missing shape of decision-grade autonomy evidence.
This chart, featured in our humanoid robotics market deck, illustrates yearly funding for humanoid robotics startups
Which humanoid has the clearest proof it can operate safely around people today?
Today, Digit shows the strongest evidence of operating safely around people. It holds the only regulator-recognized safety validation on any humanoid, an OSHA-NRTL field inspection pass at a live fulfillment site.
Agility disclosed the NRTL pass in November 2025. It is site-specific, not platform-wide, but no other humanoid has even a site-specific external regulator validation. Agility has publicly targeted ISO functional safety certification for mid-to-late 2026.
Apollo runs second. It ships with force-control architecture and a configurable Perimeter Zone, and operates alongside workers at Mercedes, GXO, and Jabil. But it has no external certification, so the safety case rests on vendor design claims.
The rest of the field lags. Figure 03 has soft coverings and foam at pinch points but no certification. Atlas ships with fenceless guarding features but no OSHA validation as of April 2026. Optimus has no external safety validation. Mantis Robotics achieved the first ISO 10218 plus ISO 13849 fenceless certification in February 2026, but Mantis is a robot arm, not a humanoid.
Which humanoid has the clearest proof it can operate with low failure rates today?
Today, Figure 02 shows the most specific reliability evidence of any humanoid. Its 1,250-hour BMW deployment tracked cycle time, placement accuracy, and interventions as explicit goals, which no other vendor has published.
Figure 02 ran 10-hour weekday shifts at BMW Spartanburg for 11 months, moving 90,000 parts. That is the most detailed operational description any humanoid vendor has published.
Digit runs second with its 100,000-tote GXO record. A Scaling Deep report cites a 98 percent success rate in Amazon Sumner pilot testing over 18 months, though this is second-hand and not customer-confirmed. Agility has not published denominators, intervention counts, or a formal error taxonomy.
The critical missing evidence across the category is decision-grade. No humanoid publishes MTBF, error taxonomies, mean-time-to-intervene, denominator-disclosed success rates, or independently audited fleet telemetry. Rodney Brooks has argued commercial humanoids need roughly 99.999 percent reliability to be broadly acceptable, and no humanoid has demonstrated that publicly.
By the way, we unpack how to turn reliability and intervention data into honest uptime forecasts in our humanoid robotics market deck.
Which humanoid has the clearest proof it can keep working over thousands of hours today?
Today, Figure 02 holds the only publicly disclosed per-deployment hours counter on any humanoid. 1,250 operating hours over 11 months at BMW Spartanburg is the specific data point.
The BMW shift pattern was 10 hours a day, five days a week, across 11 months. The data point is specific, customer-confirmed, and multi-month. No other humanoid has a comparable figure.
Digit runs second. It has operated full-time at GXO for more than a year across a growing fleet, likely exceeding Figure 02 in cumulative hours. But Agility publishes tote counts rather than explicit hours, so endurance is implied rather than stated. Optimus has internal Fremont and Giga Texas use, but Musk confirmed these are for learning, not productive tasks, so runtime quality is unclear.
Atlas commercial deployment has not started. First fleets ship to Hyundai RMAC and Google DeepMind later in 2026, with factory work not scheduled until 2028. Missing across the industry are per-unit runtime counters, maintenance intervals, and degradation curves.
This chart, featured in our humanoid robotics market deck, shows how warehouse automation has driven growth in the humanoid robotics market over time
Which humanoid has the clearest proof it can generalize across tasks today?
Today, Figure 03 shows the widest documented task breadth under one integrated AI stack. Helix 02 runs locomotion, manipulation, and balance through a single unified neural network across multiple domains.
Figure has publicly demonstrated sheet metal loading, drawer and refrigerator operation, laundry folding, dishwasher loading, and manipulation of pills, syringes, and bottle caps. A March 2026 third-party tracker cites eight distinct autonomous cleaning skills in controlled trials.
Apollo runs second with intralogistics, inspection, sorting, kitting, lineside delivery, fixture placement, and sub-assembly at Jabil. That is meaningful breadth, but it stays inside one industrial context. Digit has shown multiple warehouse tasks but all are logistics primitives.
Epoch AI's February 2026 evaluation found that across industry, household, and navigation tasks, transfer beyond trained tasks remains limited and rarely measured. That is the important caveat. Figure's range is wide in demos but not yet documented in deployed customer environments.
Which humanoid has the clearest proof it can generalize across environments today?
Today, Digit leads decisively on environment generalization. It operates productively at six named paying customers in three countries across three distinct industries.
Digit has dated productive operation at GXO Flowery Branch, Spanx, Amazon Sumner, Schaeffler plants, Toyota TMMC Woodstock, and Mercado Libre San Antonio. That spans apparel logistics, automotive manufacturing, and e-commerce fulfillment.
Apollo runs second. It operates across Mercedes-Benz Hungary, GXO lab and distribution centers, and Jabil. Respectable multi-site, multi-industry presence, but customer count and geographic spread are smaller than Digit's. Figure 02 had only one deployed site at BMW Spartanburg. Figure 03 is entering BMW Leipzig in April 2026 but has not operated productively at multiple customers.
Variables that do change across Digit sites include facility layout, product mix, conveyor systems, AMR fleets, and safety protocols. Variables that stay controlled include lighting, flooring, and largely indoor-warehouse contexts. Missing from the evidence are cross-site failure rate comparisons and independent customer attestations.
Which humanoid has the clearest proof it can replace labor cost-effectively today?
Today, no humanoid shows decision-grade evidence of replacing labor cost-effectively. Digit has the strongest indirect case through signed multi-year paid contracts, but no vendor has disclosed audited unit economics.
Agility has signed paid multi-year RaaS contracts at GXO, Toyota TMMC, and Mercado Libre. That implies customer-side ROI passed an internal threshold. Neither party discloses actual cost per hour, uptime-adjusted throughput, or margin.
Third-party reporting cites Digit at $10 to $12 per hour versus $30 per hour human labor at Amazon Sumner, though this is second-hand. Bank of America reports Western humanoid unit cost of $90,000 to $100,000 with 18-to-24-month payback.
The signed paid contracts at Digit are a meaningful signal the robot clears someone's internal hurdle rate, even without published unit economics. But 0.8-to-1.0-FTE replacement, the threshold where business cases close cleanly, is a 2027-to-2029 story for every vendor in the dataset.
Tesla's claimed future $20,000 to $30,000 price point is not delivered. 1X NEO's $499 monthly subscription relies on undisclosed teleoperator labor cost. No humanoid has published audited unit economics that demonstrate cost-effective labor substitution at wage parity.
By the way, we unpack RaaS pricing mechanics and real humanoid payback math in our humanoid robotics market deck.
As this chart shows, and as featured in our humanoid robotics market deck, search interest in where to buy robots has been rising steadily
Which humanoid can be deployed with minimal integration effort today?
Today, Digit shows the strongest cross-customer integration evidence, though no humanoid has proven true plug-and-play deployment. Agility Arc has been redeployed at six named customer sites, which is the best repeatability record in the category.
Agility Arc integrates Digit with existing WMS and AMR systems. Agility says generalization-trained policies let Digit adapt to new workflows in hours rather than weeks. The customer record complicates the plug-and-play narrative across every vendor, with typical arcs looking like this.
- Toyota TMMC's February 2026 deployment followed a year-long three-phase evaluation covering development, proof-of-technology, and live pilot
- Figure's BMW Spartanburg rollout took six months from delivery to test and ten months to full launch
Apollo runs second. It has been stood up at Mercedes, GXO, and Jabil using configurable mobility bases and swappable batteries. But every public deployment still required multi-month pilot phases, and Apollo lacks Digit's cross-customer repetition count. AgiBot reports using simulation-based validation to reduce setup time and is scaling to 100 robots by Q3 2026 at Longcheer.
Missing across the industry are disclosed customer-side engineering weeks, on-site vendor support intensity, mapping and gripper customization hours, and workflow redesign labor. Digit wins on repeatability, not on plug-and-play speed.
Which humanoid has the clearest proof it can scale without huge costs today?
Today, Unitree and AgiBot show the strongest evidence of scaling without huge costs. Between them they shipped over 10,000 humanoids in 2025 at prices as low as $13,500.
AgiBot shipped 5,168 units in 2025, a 39 percent share of the global market per Omdia's January 2026 report. Unitree shipped roughly 5,500. AgiBot hit 10,000 cumulative units by late March 2026. TrendForce projects the two combined will capture approximately 80 percent of 2026 shipments.
Western vendors lag on volume but target capability. Agility's RoboFab is designed for 10,000 Digit units per year. Figure's BotQ targets 12,000 per year initial, scaling to 100,000 over four years. Apptronik raised $935 million in February 2026. Tesla broke ground on a Giga Texas line targeting 10 million Optimus per year by 2027, an aspirational figure at aggressive pacing.
The key finding is that shipment leadership and enterprise-productive-work leadership are nearly disjoint categories. Chinese leaders ship low-cost research and consumer units. Western leaders ship enterprise-grade $100,000-plus platforms to paying customers. Missing across the industry are audited unit COGS, field support cost per unit, and yielded units delivered to paying external customers rather than internal use.
By the way, we cover manufacturing capacity, BOM curves, and cross-region unit economics in our humanoid robotics market deck.
Our methodology to build this tracker
We prioritized tier-one evidence only. That includes press releases from Agility, Figure, Apptronik, Boston Dynamics, BMW, GXO, Toyota, Mercedes-Benz, and Mercado Libre, earnings call transcripts, named executive statements, regulator filings, and reporting from The Robot Report, Bloomberg, Electrek, ASSEMBLY magazine, Epoch AI, Omdia, and TrendForce that traces to primary sources.
We framed the tracker around 11 clearest-proof-of questions rather than opinion rankings. Each question isolates one capability dimension that matters for commercial progress. We named a winner and a runner-up for each, with the specific evidence that earned the position.
We applied four filters to every credited claim. Was it dated within the last 12 months. Was it tied to a named paying customer or regulator. Did it include a quantitative denominator. Did it survive absence of disclosed teleoperation. Claims failing any filter were excluded from leader positioning.
We weighted customer disclosures more heavily than vendor claims. The distinction between clearest proof today and best positioned to win later was kept explicit throughout, because both can be true at once for different robots. Where demos were credibly reported as teleoperated, we discounted them.
