Humanoid Robot Comparison Tracker (2026)
In our humanoid robotics market deck, you will find everything you need to understand the market
Humanoid Robot Comparison Tracker
This table tracks 33 physical capability and operational questions for nine leading humanoid robots, with each cell rated Yes, Kind of, Not Really, No, or We don't know based on sourced evidence as of April 2026. Columns are ordered from most commercially advanced (left) to earliest stage (right).
The first column lists progressively harder questions, from basic locomotion and manipulation to sustained commercial performance, while each robot column shows how far that program has actually gone. If you want to dig deeper into this market, you can check out our humanoid robotics market report.
| Question | Digit (Agility) | Apollo (Apptronik) | Atlas (Boston D.) | Figure 03 | Optimus (Tesla) | NEO (1X) | IRON (XPENG) | G1 (Unitree) | GR-3 (Fourier) |
|---|---|---|---|---|---|---|---|---|---|
| Can it walk autonomously? | Yes | Yes | Yes | Yes | Kind of | No | Kind of | Yes | Kind of |
| Can it turn while walking? | Yes | Yes | Yes | Yes | Kind of | No | Kind of | Yes | Kind of |
| Can it climb stairs safely? | Yes | Kind of | Yes | We don't know | We don't know | We don't know | We don't know | Kind of | We don't know |
| Can it walk on uneven ground? | Yes | Kind of | Yes | We don't know | Kind of | No | We don't know | Kind of | We don't know |
| Can it recover from a shove? | Yes | Kind of | Yes | We don't know | We don't know | No | We don't know | Kind of | We don't know |
| Can it squat and stand? | Yes | Yes | Yes | Yes | Yes | No | Yes | Yes | Kind of |
| Can it lift ten kilograms? | Yes | Yes | Yes | Yes | Yes | Yes | We don't know | No | We don't know |
| Can it lift twenty kilograms? | Kind of | Yes | Yes | Yes | Yes | Yes | We don't know | No | We don't know |
| Can it carry weight while walking? | Yes | Yes | Yes | We don't know | We don't know | Kind of | We don't know | Kind of | We don't know |
| Can it lift from floor level? | Yes | Yes | Yes | Yes | Kind of | Kind of | Kind of | Kind of | We don't know |
| Can it push a loaded cart? | Kind of | Kind of | We don't know | We don't know | We don't know | No | We don't know | No | No |
| Can it open heavy doors? | Not Really | Kind of | We don't know | We don't know | We don't know | We don't know | We don't know | No | No |
| Can it grasp bulky objects? | Yes | Kind of | Yes | Kind of | Kind of | Kind of | Kind of | Not Really | We don't know |
| Can it grasp small objects? | Not Really | Kind of | Kind of | Yes | Yes | Yes | Yes | Kind of | Kind of |
| Can it use both hands together? | Kind of | Kind of | Yes | Yes | Kind of | Yes | Kind of | Not Really | We don't know |
| Can it place objects precisely? | Yes | Kind of | Yes | Yes | Kind of | Kind of | Kind of | Kind of | We don't know |
| Can it handle awkward loads? | Kind of | Kind of | Kind of | We don't know | We don't know | We don't know | We don't know | No | No |
| Can it keep balance under load? | Yes | Kind of | Yes | We don't know | We don't know | Kind of | We don't know | Not Really | We don't know |
| Can it repeat lifts reliably? | Yes | Kind of | We don't know | Kind of | Kind of | Kind of | We don't know | Not Really | No |
| Can it work for one hour? | Yes | Yes | Yes | Yes | Kind of | Yes | We don't know | Yes | Kind of |
| Can it work for four hours? | Yes | Yes | Yes | Kind of | We don't know | Yes | We don't know | No | Kind of |
| Can it work a full shift? | Yes | Kind of | Kind of | Not Really | No | Not Really | No | No | No |
| Can it work around humans safely? | Kind of | Yes | Kind of | Yes | We don't know | Yes | We don't know | Not Really | Yes |
| Can it finish multi-step physical tasks? | Yes | Kind of | Kind of | Kind of | Not Really | Kind of | Not Really | Not Really | Not Really |
| Can it adapt to unfamiliar objects? | Not Really | Kind of | Kind of | Kind of | Kind of | Kind of | Kind of | Not Really | Not Really |
| Can it recover from task failures? | Kind of | Kind of | We don't know | Kind of | We don't know | Kind of | We don't know | Not Really | Not Really |
| Can it work with minimal teleoperation? | Yes | Kind of | Kind of | Kind of | We don't know | No | No | Not Really | Not Really |
| Can it maintain output for days? | Yes | Kind of | Kind of | No | No | Not Really | No | No | No |
| Can it do useful work daily? | Yes | Kind of | Kind of | Kind of | Not Really | Not Really | Not Really | No | No |
| Can it do useful work for weeks? | Yes | Kind of | We don't know | Not Really | No | No | No | No | No |
| Can it match human shift reliability? | Not Really | No | No | No | No | No | No | No | No |
| Can it sustain commercial uptime? | Kind of | Not Really | No | No | No | No | No | No | No |
| Can it outperform humans on cost? | Not Really | No | No | No | No | No | No | No | No |
This market map, featured in our humanoid robotics market deck, highlights top companies and startups in the humanoid robotics market
Insights
We compared nine leading humanoid robots across 33 capability and operational questions: Agility's Digit, Apptronik's Apollo, Boston Dynamics' Atlas, Figure 03, Tesla's Optimus, 1X's NEO, XPENG's IRON, Unitree's G1, and Fourier's GR-3. Each robot was rated based on tier-one sourced evidence as of April 2026. Here is what stood out.
- Agility's Digit is the only humanoid robot generating revenue from productive commercial work as of April 2026. It moved over 100,000 totes at GXO warehouses and signed paying contracts with Toyota and Mercado Libre.
- On Tesla's Q4 2025 earnings call, CEO Elon Musk admitted Optimus units exist "primarily for learning, not productive tasks." Agility's Digit had already moved 100,000 warehouse totes when he said this.
- Every humanoid robot company claims to be building a general-purpose robot. Every actual deployment is hyper-specialized: totes for Digit, parts kits for Apollo, battery cells for Optimus, car parts for Figure 02.
- None of the nine humanoid robots we compared can match human shift reliability as of April 2026. The industry supplements narrow, repetitive tasks rather than replacing human workers at any site.
- The global fleet of commercially deployed humanoid robots doing productive work is likely under 100 units. Combined valuations of the nine companies building these robots exceed $100 billion.
- Digit passed the first OSHA-recognized safety field inspection for a commercial humanoid robot in a live warehouse. Enterprise procurement in logistics and manufacturing is gated by compliance certificates, not demo videos.
- Digit's bird-like reverse-knee legs lower its center of gravity when carrying loads in warehouses. This design improves energy efficiency and stability during repetitive walking, which is why Digit is deployed while more conventional robots stay in labs.
- Digit's end effectors are optimized specifically for plastic totes, not general-purpose manipulation. By not trying to solve dexterous hands, Agility built a robot that does one task category more reliably than any competitor.
- Digit's upcoming payload upgrade from 16 kg to 22.6 kg is a 43% increase that potentially doubles its addressable task space in warehouses. At 22.6 kg, Digit can handle loaded totes and parts kits that are currently too heavy for it.
- Digit operates for up to 8 hours on a single charge, enough for a full warehouse shift. Atlas and Apollo get 4 hours each, Figure 03 gets roughly 5 hours, and Unitree's G1 gets just 2 hours.
- Battery life is the critical bottleneck across all nine humanoid robots that receives little public attention. It is a physics problem rather than an AI problem, and physics problems do not raise venture capital.
- The humanoid robot industry converges on 20 to 25 kg payload, matching the weight of a standard loaded warehouse tote. Atlas is the lone outlier at 50 kg, priced at roughly $150,000 per unit.
- No humanoid robot company publishes maintenance data or mean-time-between-failures figures for its robot. This missing reliability metric is arguably the real bottleneck for enterprise adoption at scale.
- No humanoid robot company demonstrates what happens when a task goes wrong mid-execution. In real deployment, failure recovery is roughly 80% of the job, yet nobody shows a robot dropping an object and picking it back up.
- Figure 03's Helix AI model learned towel folding from only 80 hours of video footage. If this data efficiency generalizes to other tasks, it could be the most important technical achievement in humanoid robotics.
- 1X openly states that humans operate its NEO robot remotely while the AI learns by watching them work. This is the most transparent business model in an industry where autonomy claims often outpace actual capability.
- Every humanoid company describes a pipeline where teleoperation data trains AI until the robot becomes fully autonomous. Tesla told the same story about Full Self-Driving in 2016, and FSD still requires human supervision a decade later.
- Tesla is converting its Fremont Model S/X production lines to manufacture Optimus robots, committing $20 billion in 2026 capital expenditure. Tesla is scrapping profitable car capacity for a robot its own CEO says is not yet useful.
- Boston Dynamics committed all of its 2026 Atlas production to just two customers: parent company Hyundai and AI partner Google DeepMind. If Atlas were truly production-ready for broad deployment, Boston Dynamics would likely be selling it to any buyer willing to pay.
- Hyundai Motor Group's $26 billion U.S. investment includes a factory targeting 30,000 Atlas robots per year by 2028. This is the largest commitment of traditional manufacturing capital to humanoid robotics, dwarfing all venture funding in the sector combined.
- Automotive manufacturers fund nearly every leading humanoid robot program today. Hyundai backs Atlas, Mercedes backs Apollo, BMW backs Figure, Toyota backs Digit, and XPENG builds IRON in-house.
- Google appears as investor, AI partner, or technology provider for at least three of these nine robots: Atlas via DeepMind, Apollo via Google's direct investment, and the broader ecosystem via NVIDIA's GR00T models. Google is hedging across the entire humanoid robotics landscape.
- Chinese humanoid companies like Unitree, XPENG, and Fourier ship hardware cheap and fast, then plan to add intelligence via software updates. American companies build expensive full-stack systems before shipping, mirroring the EV industry pattern.
- Unitree shipped 5,500 humanoid G1 units in 2025, more than all Western humanoid robot companies combined. However, nearly all went to university research labs, and very few G1 units do economically productive work.
- No humanoid robot has been deployed in a consumer home for sustained daily use as of April 2026. Homes are non-standardized environments with stairs, pets, children, clutter, and variable lighting that no factory pilot can simulate.
- 1X's NEO operates at 22 dB, quieter than a typical refrigerator. No other humanoid robot company publishes noise specifications, revealing how deeply 1X has thought about home deployment compared to factory-first competitors.
- Apollo's torso can mount on bipedal legs, a wheeled base, or a stationary platform using the same core robot. This modularity hedges against the unsolved question of which mobility approach wins for different factory tasks.
- Humanoid robot prices range from $16,000 for Unitree's G1 to $250,000 for Agility's Digit. The most expensive robot is the only one generating commercial revenue from real work, suggesting capability matters more than price today.
This chart, featured in our humanoid robotics market deck, compares the main business model options for humanoid robot manufacturers
Where does Figure 03 outperform other humanoid robots?
Figure 03 outperforms the other leading humanoid robots in tactile sensitivity, home-readiness design, AI learning efficiency from minimal training data, demonstrated range of domestic tasks, and wireless charging for near-continuous operation.
On tactile sensitivity, Figure 03's fingertip sensors detect forces as low as 3 grams, enough to feel the weight of a paperclip. Atlas has "tactile-sensing hands" but has not published sensitivity thresholds. Optimus Gen 3 hands have force feedback but no gram-level resolution claims. Digit's end effectors are tote-optimized grippers with no fine touch capability. Apollo has five-digit hands but no published tactile resolution data.
On home-readiness, Figure 03 was specifically designed for household environments with soft textile coverings, a wireless floor-pad charging system, and washable garments. Atlas is a hard-shell industrial machine built for factories. Digit is built for warehouse tote handling and cannot perform household tasks. Apollo is an industrial platform designed for factory floors. Optimus is being tested only inside Tesla factories and has no home deployment timeline before 2027.
On learning efficiency, Figure 03's Helix AI model learned towel folding from only 80 hours of video footage. Optimus requires Tesla's Cortex 2.0 supercomputer with unspecified but massive data volumes for training. Atlas relies on Google DeepMind's Gemini Robotics models with undisclosed data requirements. Digit learns specific warehouse skills through extensive programming rather than video demonstration. Apollo uses NVIDIA GR00T models but has not published data efficiency metrics.
On domestic task range, Figure 03 demonstrated eight distinct autonomous cleaning skills in controlled trials in March 2026: wiping, sweeping, scrubbing, mopping, vacuuming, dusting, polishing, and organizing. Atlas focuses exclusively on industrial sequencing tasks. Digit moves totes in warehouses. Apollo delivers parts kits to assembly lines. Optimus handles battery cells in Tesla factories.
On power management, Figure 03 uses inductive foot-coil wireless charging at 2 kW with 10 Gbps mmWave data offload, allowing it to autonomously dock, charge, and resume work. Atlas uses autonomous belly-mounted battery hot-swap, which takes about 3 minutes per swap. Apollo requires a human to physically change its battery packs. Digit charges conventionally. Optimus has no published autonomous charging capability.
Where does Tesla Optimus outperform other humanoid robots?
Tesla Optimus outperforms the other leading humanoid robots in hand dexterity at production scale, manufacturing infrastructure commitment, AI training compute, vertical integration, and large language model integration for voice interaction.
On hand dexterity, Optimus Gen 3 hands have 22 degrees of freedom per hand with 50 total actuators using a tendon-driven biomimetic system, and mass production began at Fremont in January 2026. Figure 03 has 16 degrees of freedom per hand. Digit uses tote-optimized grippers, not dexterous hands. Apollo has five-digit hands but fewer published degrees of freedom. Atlas has not disclosed hand specifications. Optimus has demonstrated cracking eggs and catching thrown objects with these hands.
On manufacturing scale, Tesla is converting its Fremont Model S/X production lines to Optimus manufacturing, committing $20 billion in 2026 capital expenditure. Boston Dynamics plans 30,000 Atlas units per year by 2028 through Hyundai. Figure AI's BotQ factory targets 12,000 units per year. Agility's RoboFab in Salem has capacity for 10,000 units per year. Apptronik partners with Jabil for contract manufacturing but has not published capacity targets.
On AI training compute, Tesla's Cortex 2.0 supercomputer at Giga Texas delivers 250 MW in its first phase (April 2026) with 500 MW at full capacity by mid-2026. No other humanoid robot company has a dedicated supercomputer of this scale for robot training. Figure uses an OpenAI collaboration. Atlas gets Google DeepMind models. Apollo uses NVIDIA GR00T. Digit's Agility Arc platform handles fleet management but is not a training supercomputer.
On vertical integration, Tesla designs its own actuators, chips, sensors, and motors for Optimus. Digit uses off-the-shelf LiDAR from LIVOX. Apollo uses NVIDIA Jetson compute modules. Unitree G1 uses Intel RealSense cameras. Atlas benefits from Hyundai Mobis actuators but Boston Dynamics still designs the robot externally. Tesla's full stack should eventually produce the lowest per-unit cost among premium humanoid robots.
On voice interaction, Optimus integrates Grok, xAI's frontier large language model, for real-time voice commands and clarifying questions. Figure 03 has speech-to-speech conversation. NEO has an onboard LLM. Atlas has not demonstrated conversational capability. Apollo has LED status displays but no published voice interaction system.
Where does Agility Digit outperform other humanoid robots?
Agility's Digit outperforms the other leading humanoid robots in commercial deployment maturity, battery endurance, fleet management software, OSHA safety validation, and proven real-world reliability over extended periods.
On commercial deployment, Digit is the only humanoid robot generating revenue from productive commercial work. It moved over 100,000 totes at GXO's Flowery Branch warehouse and Toyota Canada signed a commercial agreement for seven units at its RAV4 plant. Atlas has not shipped to any external customer. Optimus is not doing useful work per Musk's own admission. Figure 03 is in a single pilot at BMW. Apollo is in controlled pilot deployments at Mercedes-Benz.
On battery life, Digit operates for up to 8 hours on a single charge, enough for a full warehouse shift. Atlas gets 4 hours per battery. Apollo gets 4 hours per swappable pack. Figure 03 gets approximately 5 hours. NEO gets approximately 4 hours. Unitree G1 gets 2 hours. No other humanoid robot can cover a standard 8-hour work shift without recharging or swapping batteries.
On fleet management, Agility Arc is the first and only humanoid fleet management platform proven in commercial deployment, integrating with warehouse management systems and autonomous mobile robots from MiR and Zebra. Atlas has no fleet management system deployed. Optimus has fleet ambitions but nothing operational. Apollo has no published fleet platform. Figure has no equivalent system. This software layer makes Digit deployable in existing warehouse infrastructure.
On safety validation, Digit is the first commercial humanoid to pass an OSHA-recognized safety field inspection in a live warehouse. Atlas has fenceless guarding features but no OSHA validation. Optimus safety is unvalidated by any external body. Apollo was designed for human co-working but lacks external safety certification. Figure 03 has soft coverings but is not safety-certified. This certification is a prerequisite for enterprise adoption.
On sustained reliability, Digit has worked full-time at GXO for over a year, handling live production workflows with genuine throughput requirements. It adapts to workflow fluctuations by stacking totes aside when downstream stations are backed up. Atlas had its first public demo in January 2026. Optimus units are learning, not working. Apollo and Figure are in controlled pilot conditions. No other robot has demonstrated this duration of continuous real-world operation.
This chart, featured in our humanoid robotics market deck, shows how Agility Robotics is capturing share in humanoid robotics
Where does Boston Dynamics Atlas outperform other humanoid robots?
Boston Dynamics' Atlas outperforms the other leading humanoid robots in raw physical strength, degrees of freedom and range of motion, operating temperature range, autonomous battery self-swap for continuous operation, and depth of its Google DeepMind AI partnership.
On physical strength, Atlas lifts 50 kg (110 lbs) with a reach of up to 7.5 feet. Digit carries 16 kg, upgrading to 22.6 kg. Apollo carries 25 kg. Figure 03 carries 20 kg. Optimus carries 20 kg. Atlas can handle genuinely heavy industrial parts that no other humanoid robot on this list can touch, which is critical for automotive assembly where heavy components are standard.
On degrees of freedom and range of motion, Atlas has 56 degrees of freedom with 360-degree rotation at hips, waist, and neck joints. Figure 03 has 44 degrees of freedom. Apollo has 71 degrees of freedom but many are in finger joints, not the core body. Optimus Gen 2 body has fewer core-body degrees of freedom than Atlas. Digit has far fewer total degrees of freedom. Atlas's superhuman joint rotation lets it approach tasks from angles impossible for all competitors.
On operating temperature, Atlas works from -4 F to 104 F (-20 C to 40 C), the broadest published range among these nine robots. Most competitors do not publish temperature specs, suggesting they are limited to indoor controlled environments. Digit, Apollo, Figure 03, and Optimus have no published operating temperature ranges. Atlas can work in cold storage, outdoor loading docks, and hot factory floors.
On power management, Atlas autonomously navigates to swap its own belly-mounted battery in about 3 minutes, then returns to work without human intervention. Apollo has hot-swappable batteries but requires a human to change them. Digit charges conventionally. Figure 03 wireless-charges but must park on a floor pad. Optimus has no published battery swap capability. Atlas's self-swap is the most autonomous power management system among all nine robots.
On AI partnership, Atlas is receiving Gemini Robotics foundation models directly from Google DeepMind, with DeepMind receiving physical Atlas robots for bidirectional integration work. Apollo gets DeepMind models through Google's investment, but not a dedicated co-development program. Figure developed its Helix AI internally. Optimus uses xAI's Grok. Digit uses Agility's own software. The dedicated Atlas-DeepMind integration could produce the most intelligent industrial robot in this group.
Where does Apptronik Apollo outperform other humanoid robots?
Apptronik's Apollo outperforms the other leading humanoid robots in hot-swappable battery runtime per day, human-centric safety design, platform modularity, breadth of enterprise partnerships, and space-grade engineering heritage from NASA.
On daily runtime, Apollo's battery packs swap in seconds, enabling up to 22 hours of operation per day by rotating six 4-hour packs. Atlas self-swaps but takes about 3 minutes per swap and gets 4 hours per battery. Digit gets 8 hours total but then must charge for significant downtime. Figure 03's wireless charging still creates downtime. Optimus has no swappable battery. For factories running two or three shifts, Apollo's swap system delivers near-zero downtime with just a shelf of charged batteries.
On safety design, Apollo was built from inception for safe human co-working with force control architecture, configurable safety zones that detect, slow, and stop at different ranges, and LED status displays on head and chest. Digit currently operates in segregated zones away from humans. Atlas has fenceless guarding but is primarily an industrial machine. Optimus safety around humans is unvalidated. Figure 03 has soft coverings but no published safety architecture. Apollo's force-sensing joints respond in real time to contact, making it the safest-feeling robot to work alongside.
On modularity, Apollo's torso and arms can mount on bipedal legs, a wheeled base, or a stationary platform. Atlas is bipedal legs only. Digit is bipedal legs only. Figure 03 is bipedal legs only. Optimus is bipedal legs only. This means a single Apollo purchase can serve multiple use cases: walking in a warehouse, rolling on a factory line, or fixed at a workstation, reducing total cost of ownership.
On enterprise partnerships, Apollo has simultaneous active engagements with Mercedes-Benz in automotive, GXO in logistics, Jabil in electronics manufacturing, Google in AI, and NVIDIA in compute. Digit has GXO and Toyota. Atlas has Hyundai and DeepMind. Figure has BMW. Optimus is internal to Tesla with no external partners. Apollo's multi-industry validation across automotive, logistics, and electronics manufacturing demonstrates broader applicability than any single competitor.
On engineering heritage, Apptronik's team built NASA's Valkyrie (R5) robot, designed for space exploration where failure tolerance is zero. Digit comes from Oregon State University walking robot research. Atlas comes from DARPA military robotics challenges. Figure AI is a three-year-old startup. Tesla Optimus comes from automotive engineering. Apollo's space-grade engineering DNA translates to industrial reliability expectations that other robots have not been tested against.
This chart, featured in our humanoid robotics market deck, illustrates yearly funding for humanoid robotics startups
This tracker is a live research tool we maintain to compare nine leading humanoid robots on physical capability and operational readiness. It is updated as new evidence emerges.
The tracker is focused on one question: what can each humanoid robot actually do today, and how does it compare to the others? It is not designed to score potential, funding, or future roadmaps.
To be included, a company must have a publicly announced full-body humanoid robot program with observable evidence of physical operation. Upper-body systems, wheeled service robots, exoskeletons, and robot heads are excluded. 1X NEO is included despite using wheels because it has a full humanoid form factor.
The 33 questions are ordered by increasing difficulty. Early rows cover basic locomotion and manipulation. Later rows cover sustained useful work, endurance, and commercial viability. This structure makes it visually obvious where each robot's demonstrated capability runs out.
Each cell is rated Yes, Kind of, Not Really, No, or We don't know. "Yes" means credible sourced evidence exists. "Kind of" means partial or inconsistent evidence exists. "Not Really" means the robot struggles or mostly fails. "No" means evidence confirms it cannot. "We don't know" means no credible public data was found either way.
We rely on direct and credible sources: company press releases, product pages, customer announcements, SEC filings, IPO prospectuses, executive statements on earnings calls, and reporting from publications like TIME, TechCrunch, IEEE Spectrum, Engadget, and The Robot Report. Demo videos and unverified social media claims are not treated as strong evidence.
When evidence is partial or hard to verify, we score conservatively. The tracker should be read as directional, not absolute. Columns are ordered from most commercially advanced (left) to earliest stage (right). You can find more details and additional analysis in our humanoid robotics market report.
This chart, featured in our humanoid robotics market deck, shows how warehouse automation has driven growth in the humanoid robotics market over time
Related blog posts
- The startups that have raised the most funding in the humanoid robotics market
- The most highly valued startups in the humanoid robotics market
- The full range of business models in the humanoid robotics market
- How close is Tesla Optimus to being deployed at scale?
- Which humanoid robotics companies seem closest to scale?
- What are the latest production milestones from humanoid robot companies?
