Humanoid robot applications
Last reviewed
Jun 10, 2026
Sources
56 citations
Review status
Source-backed
Revision
v5 · 6,314 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 10, 2026
Sources
56 citations
Review status
Source-backed
Revision
v5 · 6,314 words
Add missing citations, update stale details, or suggest a clearer explanation.
Humanoid robot applications span a growing range of industries and use cases where the human-like form factor provides distinct advantages over traditional fixed-base or wheeled robots. Because humanoid robots are designed to operate in environments built for people, they can navigate stairs, use human tools, and interact naturally with workers and customers. As of 2025, deployments range from commercially operational factory lines to early-stage research prototypes for space exploration.
Not all humanoid robot applications have reached the same level of maturity. The following readiness framework helps categorize the current state of each application area.
| Readiness level | Description | Examples |
|---|---|---|
| Commercially Deployed | Robots operating in real-world production or commercial settings with paying customers | Manufacturing lines, logistics warehouses, education programs, robotics competitions |
| Pilot Programs | Robots undergoing structured trials in controlled real-world environments with limited scope | Healthcare facilities, hospitality venues, museums, teleoperation demonstrations |
| In Development | Functional prototypes demonstrated but not yet deployed outside labs or controlled demos | Consumer home robots, personal assistants |
| Research Stage | Early-stage exploration with academic or government research partners | Hazardous environment operations, space exploration |
The pace of advancement has accelerated significantly since 2023, with several applications moving from pilot programs to commercial deployment within 12 to 18 months.
Readiness: Commercially Deployed
Manufacturing represents the most mature commercial application for humanoid robots. The automotive sector has been the earliest large-scale adopter, driven by labor shortages, ergonomic concerns for human workers, and the need for flexible automation that can adapt to model changeovers without expensive retooling.
Figure 02 robots have been deployed at BMW's Spartanburg plant in South Carolina, where they handle metal sheet placement for welding operations.[1] The robots work alongside human workers on the production line, picking up metal body panels and positioning them precisely for robotic welding arms. This deployment began in early 2024 and marked one of the first instances of a general-purpose humanoid robot performing sustained production-line work in a major automotive facility.
Apptronik Apollo robots entered service at Mercedes-Benz facilities in Germany and Hungary, focusing on intralogistics tasks such as transporting parts bins between storage areas and assembly stations.[2] Apollo's design emphasizes payload capacity and battery endurance, which are critical for the repetitive material transport cycles found in automotive logistics.
UBTECH Walker S1 has achieved the broadest manufacturing deployment among humanoid robots. The Walker S1 operates at factories run by BYD, Geely, Foxconn, and FAW Hongqi, with UBTECH reporting over 500 orders as of early 2025.[3] In component sorting tasks, the Walker S1 achieves 99.7% precision, which approaches the accuracy required for quality-critical automotive parts handling.[3]
XPeng Iron robots have been integrated into XPeng's Guangzhou factory, where they participate in production of the P7+ electric sedan.[4] XPeng's approach leverages the company's existing expertise in AI and sensor systems from its electric vehicle business.
Humanoid robots in manufacturing currently perform three primary task categories:
| Task category | Description | Example deployments |
|---|---|---|
| Assembly and packaging | Placing components, fastening parts, packaging finished goods | Figure 02 at BMW (sheet metal placement) |
| Material handling | Transporting parts, loading/unloading bins, feeding production lines | Apollo at Mercedes-Benz (bin transport) |
| Quality inspection | Visual and tactile inspection of components, sorting by quality grade | Walker S1 at BYD (99.7% sorting precision) |
Fine manipulation remains a significant gap. Tasks requiring dexterous finger control, such as inserting small fasteners or routing wiring harnesses, still exceed the capabilities of most humanoid robot hands. Production line timing presents another constraint: humanoid robots must match the cycle times of existing processes (often 60 to 90 seconds per station in automotive assembly) or risk becoming bottlenecks. Cost justification against traditional industrial automation is also a persistent challenge, as a fixed robotic arm can often perform a single repetitive task more cheaply than a humanoid, though the humanoid's ability to be redeployed across multiple tasks provides a flexibility advantage.
In February 2026, BMW reported that the Figure 02 pilot at Spartanburg had concluded after roughly 1,250 operating hours of ten-hour daily shifts over 11 months, during which the robots loaded more than 90,000 sheet metal parts and supported production of over 30,000 BMW X3 vehicles.[28] BMW simultaneously announced its first humanoid deployment in Europe: the Hexagon AEON robot is scheduled to begin work at BMW Plant Leipzig in summer 2026, assembling high-voltage batteries and supporting component manufacturing, coordinated through a new Center of Competence for Physical AI in Production.[28]
UBTECH moved from single-robot pilots toward fleet operations. In early 2025 it ran what it described as the world's first multi-robot collaborative training program at Zeekr's 5G Intelligent Factory, where dozens of Walker S1 robots practiced coordinated sorting, handling, and precision assembly under its BrainNet framework for swarm intelligence.[29] In July 2025 the company introduced the UBTECH Walker S2, which can swap its own battery in about three minutes to support continuous operation.[30] Walker S2 entered mass production in November 2025 with a first batch of several hundred units and a 500-unit delivery target for the year; UBTECH reported cumulative Walker-series orders above 800 million yuan (about $112 million) since early 2025 from customers including BYD, Dongfeng Liuzhou Motor, Geely, FAW-Volkswagen Qingdao, Audi FAW, BAIC New Energy, Foxconn, and SF Express, and set production capacity targets of 5,000 robots in 2026 and 10,000 in 2027.[31]
Other manufacturers followed. Reuters reported in May 2025 that Foxconn and NVIDIA planned to put humanoid robots powered by NVIDIA Isaac GR00T models on the production line of Foxconn's Houston plant that builds NVIDIA AI servers, targeting the first quarter of 2026.[32] At CES in January 2026, Boston Dynamics said commercial production of the electric Atlas had begun and that its entire 2026 output was committed to Hyundai's Robotics Metaplant Application Center and to Google DeepMind, which is integrating Gemini Robotics foundation models into the robot; Hyundai also announced plans for a United States robotics factory with capacity for 30,000 robots per year.[33] Tesla Optimus remained at the pilot stage: Tesla planned to host a larger Optimus production line at its Fremont factory in space previously used for Model S and Model X production, and in April 2026 Elon Musk said volume production of the third-generation design was expected to begin around late July or August 2026, with slow initial output.[34]
Readiness: Commercially Deployed
Logistics warehouses present an ideal deployment environment for humanoid robots because they are structured yet dynamic, with standardized shelving and bin systems but constantly changing inventory configurations.
Agility Robotics Digit achieved the earliest large-scale logistics deployment among humanoid robots. Working at GXO Logistics facilities handling Spanx products, Digit robots processed over 10,000 orders by August 2024.[5] The robots perform pick-and-place operations, moving products from shelving to shipping containers. Digit handles payloads up to 35 lbs (15.9 kg), which covers the majority of e-commerce package weights.
Amazon has deployed Digit robots at warehouses near Seattle for tote consolidation, where the robots move plastic totes between conveyor systems and storage racks.[6] This deployment focuses on tasks that are ergonomically challenging for human workers, particularly repeated bending and lifting at floor level.
Apptronik Apollo robots also operate in Mercedes-Benz logistics centers for bin handling tasks, complementing their manufacturing deployments at the same facilities.[2]
Galbot G1 robots have been deployed at Meituan facilities in China, performing 24-hour inventory replenishment cycles.[7] The robots restock shelves during overnight hours when human workers are not present, demonstrating a key advantage of humanoid robots: the ability to work continuously across all shifts without fatigue.
Handling varied and deformable objects (such as polybags, soft goods, and irregularly shaped items) remains difficult for current grippers and perception systems. Operating at speeds matching experienced human pickers (typically 100 to 150 picks per hour) while maintaining safety around human coworkers requires sophisticated motion planning and force limitation systems.
By late 2025, Agility Robotics reported that Digit had moved more than 100,000 totes at GXO's fulfillment center in Flowery Branch, Georgia, under the industry's first humanoid robots-as-a-service agreement.[35] In November 2024, German motion-technology supplier Schaeffler made a strategic investment in Agility Robotics and said it saw potential to deploy a significant number of humanoids across its global network of roughly 100 plants by 2030.[36]
Humanoid logistics work also began extending beyond the warehouse toward last-mile delivery. In June 2025, The Information reported that Amazon had built an indoor obstacle course, internally called a "humanoid park," at one of its San Francisco offices to test humanoid robots for carrying packages from Rivian electric delivery vans to doorsteps, using third-party hardware (including robots from Unitree) running Amazon-developed software.[37]
Readiness: Pilot Programs
Healthcare applications leverage the humanoid form factor for patient comfort and trust, as people tend to respond more positively to robots with human-like proportions when receiving physical assistance or emotional support.
CloudMinds has deployed humanoid service robots across elderly care facilities in China, with the company reporting that 68% of its orders come from national elderly care facilities.[8] These robots provide companionship, medication reminders, fall detection, and communication links to family members and medical staff.
Fourier GR-1 robots have been deployed in pilot programs at China Construction Bank wellness centers and SAIC-GM employee health facilities, focusing on rehabilitation assistance.[9] The GR-1 guides patients through physical therapy exercises using force feedback to ensure proper form and appropriate resistance levels.
The RIBA (Robot for Interactive Body Assistance) system, developed by RIKEN, demonstrated patient transfer capabilities, lifting patients from beds to wheelchairs.[10] While RIBA itself is no longer in active deployment, it established foundational research for patient handling that continues to influence current healthcare robotics development.
| Capability | Description | Technical requirements |
|---|---|---|
| Patient companionship | Conversation, activity engagement, emotional monitoring | Natural language processing, emotion recognition |
| Sit-to-stand assistance | Physical support during transfers between bed, chair, and standing | Force control, balance estimation, 50+ kg payload |
| Rehabilitation exercises | Guided movement with adjustable resistance and form correction | Force feedback, joint angle sensing, clinical protocols |
| Medication management | Reminders, dispensing, compliance tracking | Facial recognition, schedule management, secure storage |
Safety certification for physical human-robot interaction in healthcare settings requires meeting stringent standards (such as IEC 80601 for medical electrical equipment). Patient trust varies significantly across demographics and cultures, and many elderly patients initially resist robotic assistance. Handling unpredictable patient behavior, including sudden movements during transfers or confused responses, demands robust real-time safety systems.
In August 2025, Fourier unveiled the GR-3, a 165 cm, 71 kg "care-bot" companion humanoid with 55 degrees of freedom, soft warm-touch surfaces, and a full-sensory interaction system combining auditory, visual, and tactile modules. Fourier positioned the robot for nursing homes, rehabilitation centers, and other institutional buyers at a price above 200,000 yuan (about $27,500),[38] and gave the GR-3 its international debut at CES 2026.[39]
Standards work also advanced. In January 2025 the International Electrotechnical Commission published IEC 63310, the first international standard for elderly care (active assisted living) robots, which sets functional performance criteria covering health monitoring, emergency detection, communication support, household assistance, and mobility support in connected home environments.[40]
Readiness: Research Stage
Hazardous environment applications capitalize on the fundamental advantage of sending robots where humans cannot safely go. The humanoid form is valuable in these settings because hazardous facilities were originally designed for human workers, with human-scale doorways, ladders, control panels, and tool interfaces.
Boston Dynamics has partnered with nuclear facility operators to test robot inspections in radioactive environments, where robots can enter contaminated zones to take readings, capture imagery, and manipulate valves without exposing human workers to radiation.
NASA maintains active partnerships with multiple humanoid robot developers for space exploration applications. The Valkyrie (R5) platform has been used to study how humanoid robots could maintain spacecraft systems, perform extravehicular repairs, and prepare habitats on the Moon or Mars before human crew arrival.[11]
NASA has also applied Valkyrie to terrestrial hazardous-environment work. Under a reimbursable Space Act Agreement announced in 2023, the robot was shipped to Woodside Energy in Perth, Australia, to help develop remote mobile dexterous manipulation capabilities for the caretaking of uncrewed and offshore energy facilities, with the resulting software and operational data intended to inform robotic caretakers for Artemis-era lunar worksites and habitats.[41]
| Environment | Hazard type | Humanoid advantage |
|---|---|---|
| Nuclear facilities | Ionizing radiation | Can operate in zones lethal to humans within minutes |
| Chemical plants | Toxic exposure, explosion risk | No respiratory or dermal exposure risk |
| Disaster zones | Structural collapse, fire, flooding | Can navigate rubble and debris on bipedal legs |
| Deep ocean | Extreme pressure, no breathable atmosphere | Bipedal manipulation of subsea equipment designed for divers |
| Space | Vacuum, radiation, microgravity | Can use human-designed tools and interfaces on spacecraft |
Hardening electronics and actuators against radiation, extreme temperatures, and pressure differentials adds significant cost and weight. Communication latency in space applications (up to 24 minutes for Mars) requires high levels of autonomy. These applications remain in the research stage primarily because the engineering requirements for environmental hardening add years to development timelines.
Readiness: Pilot Programs
Teleoperation allows a human operator to control a humanoid robot remotely, effectively projecting their physical presence to a distant location. This approach combines human judgment and dexterity with the robot's ability to operate in dangerous or distant environments.
NASA's Valkyrie platform has been tested for improvised explosive device (IED) response scenarios, where a human operator controls the robot from a safe distance to inspect and handle suspicious objects.
The TELESAR VI system, developed at Keio University, provides full haptic feedback to the operator through a glove-based interface. The operator can feel the shape, texture, and temperature of objects the robot touches, enabling delicate manipulation tasks that require tactile sensing.[12]
Boston Dynamics Atlas served as a primary platform during the DARPA Robotics Challenge (2012 to 2015), where teams demonstrated teleoperated driving, door opening, valve turning, and debris clearing.[25]
GITAI G1 robots have been tested in partnership with JAXA (Japan Aerospace Exploration Agency) for spacecraft maintenance tasks, with the operator on Earth controlling the robot aboard the International Space Station or future lunar facilities.[13]
Teleoperation has also become a bridge technology for consumer humanoids. The NEO home robot from 1X Technologies, opened for pre-order in late October 2025, performs some chores autonomously but lets owners schedule remote 1X operators, which the company calls Experts, to guide the robot through unfamiliar tasks while generating training data for future autonomy; privacy controls include scheduled sessions, customer-defined no-go zones, and automatic blurring of faces.[42]
| Capability | Technology | Maturity |
|---|---|---|
| Real-time motion retargeting | Inverse kinematics mapping from operator to robot | Mature |
| VR visual feedback | Stereoscopic cameras with head tracking | Mature |
| Haptic feedback | Force/torque sensors with tactile actuator gloves | Emerging |
| Whole-body exoskeleton control | Full-body motion capture suit with force feedback | Research stage |
Communication latency is the primary constraint. Even small delays (over 100 milliseconds) degrade operator performance significantly. Bandwidth limitations affect visual feedback quality, particularly in remote or space-based operations. Operator fatigue during extended teleoperation sessions is also a concern, as controlling a humanoid robot requires sustained concentration.
Readiness: Pilot Programs
The hospitality industry values humanoid robots for their ability to create memorable guest experiences while handling routine service tasks that free human staff for higher-value interactions.
IntBot's Nylo robot was showcased at CES 2025, demonstrating hotel concierge capabilities including guest check-in assistance, direction-giving, and room service delivery. The robot features natural conversation abilities and can recognize returning guests.[14]
Service robots with humanoid upper bodies have been deployed in hotels and restaurants across Asia, particularly in Japan, South Korea, and China. These robots take orders, deliver food and beverages to tables, and provide multilingual information to international guests.
Humanoid service robots currently handle customer greeting and interaction, order taking and transmission to kitchen or service systems, item delivery along predefined routes, and multilingual conversation for international venues. The humanoid form factor provides a more approachable interface than screen-based kiosks or wheeled delivery robots.
Handling unexpected situations (spills, guest complaints, unusual requests) requires either advanced AI reasoning or seamless handoff to human staff. The cost of humanoid robots relative to simpler service automation (tablets for ordering, conveyor systems for delivery) makes the business case dependent on the novelty and customer experience value.
At its AI Day on November 5, 2025, XPeng unveiled a next-generation IRON humanoid featuring a biomimetic spine and muscle system, flexible outer skin, a solid-state battery, and three of the company's Turing AI chips delivering 2,250 TOPS of computing performance. XPeng said large-scale production would begin by the end of 2026, with initial deployments as interactive service robots in museums, auto showrooms, and shopping centers.[43]
In retail, Galbot extended its G1 from warehouse work into customer-facing stores. Its unstaffed Galbot Store format, in which the robot retrieves products from shelves and serves customers around the clock, had reached more than 30 cities across China by December 2025, when the company raised over $300 million at a $3 billion valuation and cited manufacturing partnerships with CATL, Bosch, Toyota, and Hyundai.[44]
Readiness: In Development
Home deployment represents the ultimate test for humanoid robots, as residential environments are the most unstructured and variable settings these machines will encounter.
1X Technologies NEO Gamma is planned for testing in hundreds of homes by the end of 2025, making it potentially the first humanoid robot to enter residential environments at meaningful scale.[15] The NEO platform focuses on safe human interaction through compliant actuators and lightweight construction.
Figure 03 has been demonstrated performing home tasks in controlled settings, including kitchen cleanup and object organization.[16] Figure's approach emphasizes large language model integration for understanding natural language instructions about household tasks.
Neura Robotics 4NE-1 targets household tasks such as ironing, folding, and general tidying.[17] The robot uses a combination of vision-based planning and learned manipulation skills.
Clone Robotics Alpha has announced a limited series of 279 units targeting early-adopter home use, with the robot's musculoskeletal actuation system designed to provide more natural and safe movement in close proximity to people.[18]
| Task | Difficulty level | Key technical challenge |
|---|---|---|
| Laundry folding | Very high | Deformable object manipulation, fabric state estimation |
| Dishwasher loading | High | Varied object geometries, fragile item handling |
| Room tidying | High | Scene understanding, object categorization, placement logic |
| Item fetching | Medium | Navigation, object recognition, grasp planning |
| Cooking assistance | Very high | Tool use, heat management, multi-step sequencing |
The home environment presents the highest motion control requirements of any humanoid application. Deformable objects (fabrics, soft foods, flexible packaging) defy the rigid-body assumptions underlying most robot manipulation algorithms. Cluttered environments with varied lighting, reflective surfaces, and transparent objects challenge perception systems. Safety around children and pets demands extremely reliable collision detection and force limitation, as contact with vulnerable individuals could cause serious injury. The home also demands quiet operation, as actuator noise that is acceptable in a factory would be disruptive in a living space.
1X opened consumer pre-orders for NEO in late October 2025 at $20,000, or a $499 per month subscription, with a $200 deposit, and said deliveries would begin in the United States in 2026 before expanding to other markets in 2027.[42] In April 2026 the company opened a NEO factory in Hayward, California, which it describes as America's first vertically integrated humanoid robot factory, to support consumer shipments planned for 2026.[45]
Figure formally unveiled Figure 03 on October 9, 2025 as a home-oriented design built around its Helix vision-language-action model, with soft textile coverings, multi-density foam over pinch points, fingertip tactile sensors able to detect forces of about three grams, and 2 kW wireless inductive charging through coils in the feet.[46] The robot is manufactured at Figure AI's BotQ facility, which the company says has a first-generation capacity of up to 12,000 robots per year toward a goal of 100,000 robots over four years;[46] in spring 2026 Figure said BotQ output had risen from one robot per day to roughly one robot per hour.[47]
Readiness: Commercially Deployed
Education and research represent one of the longest-established application areas for humanoid robots, with platforms dating back to the early 2000s.
SoftBank Robotics NAO has been used as an educational tutor in schools worldwide for over a decade.[19] NAO teaches programming concepts, foreign languages, and STEM subjects through interactive exercises. Its small size (58 cm) and friendly appearance make it particularly effective with younger students.
UBTECH operates AI Education programs that have trained over 500,000 students globally in robotics and AI concepts.[20] The programs use UBTECH's range of humanoid robots as hands-on learning platforms, covering topics from basic programming to advanced machine learning.
Unitree H1 and Unitree H2 have become popular research platforms at universities worldwide, offering capable hardware at price points accessible to academic budgets. Their open software architecture allows researchers to implement and test new locomotion, manipulation, and perception algorithms.
Booster Robotics T1 positions itself as a developer platform priced between $20,000 and $30,000, with full API access and ROS2 compatibility.[21] This price point makes it accessible to university research labs and robotics startups that cannot afford platforms costing hundreds of thousands of dollars.
Fourier Intelligence, Unitree, and Booster Robotics have established partnerships with the RoboCup Federation to provide standardized humanoid platforms for competition and research, helping to create a common hardware baseline for comparing algorithmic approaches.
University labs use humanoid robots to advance research in bipedal locomotion, whole-body manipulation, reinforcement learning for motor control, computer vision for unstructured environments, human-robot interaction, and natural language processing for task understanding.
Readiness: Pilot Programs
Public space deployments test humanoid robots' ability to interact with untrained members of the general public, providing valuable data on human-robot interaction at scale.
Honda ASIMO was deployed at the Miraikan (National Museum of Emerging Science and Innovation) in Tokyo, where a study of 231 participants found that 94.74% wanted to interact with ASIMO again.[23] The average interaction duration was 9 minutes, suggesting sustained engagement rather than brief novelty-driven encounters.[23]
TritonBot has been deployed as a museum guide, providing information about exhibits and navigating visitors through galleries. The robot uses natural language understanding to answer questions about displayed items.
Pepper, developed by SoftBank Robotics, has been deployed in museums and exhibitions worldwide as a greeter and information assistant. Pepper's emotion recognition capabilities allow it to adjust its interaction style based on the visitor's apparent mood.
Robovie, developed by ATR in Japan, has been deployed in shopping malls with RFID-based visitor identification. The robot recognizes returning visitors via RFID tags and personalizes its greetings and recommendations based on past interactions.
Research on proactive engagement (where the robot initiates interaction rather than waiting to be approached) has achieved 95.3% accuracy in attention estimation, allowing robots to identify visitors who are looking at them or appear interested in interaction.
Public space robots must handle extremely diverse interaction partners, from young children to elderly visitors, across languages and cultural norms.[24] Vandalism and rough handling by unsupervised visitors present durability concerns. Navigation in crowded spaces with unpredictable pedestrian flow requires robust path planning and obstacle avoidance.
Readiness: Commercially Deployed
Robotics competitions serve as proving grounds for humanoid robot capabilities and as catalysts for research advancement. They establish standardized benchmarks that allow meaningful comparison of different approaches.
| Competition | Founded | Format | Notable achievement |
|---|---|---|---|
| RoboCup | 1997 (annual) | Autonomous soccer, rescue, service tasks | Goal: beat human FIFA World Cup champions by 2050 |
| DARPA Robotics Challenge | 2012 (ended 2015) | Disaster response tasks (driving, door opening, valve turning) | Won by Team KAIST with DRC-HUBO (8 tasks in 44 min 28 sec) |
| FIRA | 1997 (annual) | Robot soccer, humanoid events | Parallel development to RoboCup with different rule sets |
| HuroCup | Annual | Pentathlon (sprint, obstacle run, lifting, climbing, marathon) | Tests athletic capabilities across multiple events |
| CYBATHLON | 2016 (biennial) | Assistive technology for people with disabilities | Bridges competition robotics and real-world assistive needs |
| World Humanoid Robot Games | 2025 (first edition) | Track and field, soccer, dance, and applied task events for humanoid robots | 280 teams from 16 countries competed in Beijing in August 2025; Unitree won four golds [48] |
| Beijing humanoid robot half marathon | 2025 (first edition) | 21.1 km road race for humanoid robots held alongside a human half marathon | Tiangong Ultra won the inaugural race in 2:40:42 [49] |
The DARPA Robotics Challenge (2012 to 2015) is particularly significant for its role in advancing full-sized humanoid capabilities. The competition required robots to drive vehicles, open doors, cut through walls, turn valves, and climb stairs, all tasks relevant to disaster response. Team KAIST won the final competition in 2015 with their DRC-HUBO robot, completing all eight tasks in 44 minutes and 28 seconds.[25]
RoboCup's long-term goal of fielding a team of autonomous humanoid robots capable of defeating the human FIFA World Cup champions by 2050 drives sustained research in bipedal locomotion, real-time perception, multi-agent coordination, and robust decision-making under uncertainty.[26]
CYBATHLON stands apart from other competitions by focusing on assistive technology for people with physical disabilities. The competition features events using powered exoskeletons, prosthetic limbs, and brain-computer interfaces, directly connecting humanoid robotics research to practical assistive applications.
Two large-scale events in China expanded competitive humanoid robotics in 2025. On April 19, 2025, Beijing hosted the world's first humanoid robot half marathon, in which robots ran a separated course alongside roughly 12,000 human runners; Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center, won in 2 hours 40 minutes and 42 seconds,[49] and only six of the 21 robot entrants finished the course.[50] Four months later, the first World Humanoid Robot Games were held from August 14 to 17, 2025 at Beijing's National Speed Skating Oval, drawing 280 teams from 16 countries; Unitree won four track golds (the 1,500 m, 400 m, 4x100 m relay, and 100 m hurdles), with its H1 robot finishing the 1,500 m in 6 minutes 34 seconds.[48]
The decision to deploy a humanoid robot rather than a traditional industrial robot depends on the specific requirements of the application. The following comparison highlights the key tradeoffs.
| Factor | Traditional industrial robots | Humanoid robots |
|---|---|---|
| Mobility | Fixed base or rail-mounted; limited to one workstation | Bipedal locomotion; can move between workstations, navigate stairs, and access any area designed for humans |
| Flexibility | Task-specific end effectors; reprogramming required for new tasks | General-purpose hands; can adapt to multiple tasks with software updates |
| Infrastructure | Requires safety cages, custom fixtures, dedicated floor space | Operates in existing human workspaces with minimal modification |
| Precision | Sub-millimeter repeatability (typically ±0.02 to ±0.1 mm) | Centimeter-level precision (typically ±5 to ±20 mm); improving with better actuators |
| Cost | $50,000 to $400,000 per unit (mature supply chain) | $30,000 to $150,000+ per unit (costs declining rapidly as production scales) |
| Speed | Very fast cycle times for repetitive tasks | Slower than dedicated automation; comparable to human worker speeds |
| Best for | High-volume, repetitive, precision tasks at fixed locations | Variable tasks, multi-station workflows, environments designed for humans |
The key insight is that humanoid robots do not replace industrial robots in their areas of strength. Instead, humanoid robots address the large category of tasks that were previously impossible to automate because they require mobility, adaptability, and operation in human-designed spaces. Industry analysts estimate that traditional automation addresses roughly 25% of manual tasks in manufacturing and logistics, while humanoid robots could eventually address a significant portion of the remaining 75%.
Several challenges affect humanoid robot deployments across all application areas.
Battery life and energy efficiency. Most current humanoid robots operate for 2 to 4 hours on a single charge. Manufacturing and logistics applications requiring 8-hour or 24-hour operation need either rapid battery swapping, autonomous charging stations, or significant improvements in energy storage density.
Regulatory frameworks. Existing industrial safety standards (ISO 10218, ISO/TS 15066 for collaborative robots) were not designed for mobile humanoid robots that move between workstations and interact closely with humans. New standards and certification processes are being developed but remain incomplete.
Software and AI maturity. While hardware capabilities have advanced rapidly, the AI systems for understanding complex environments, planning multi-step tasks, and recovering from errors are still maturing. Foundation models and large language models are accelerating progress in task understanding, but reliable autonomous operation in unstructured environments remains an open research problem.
Workforce integration. Deploying humanoid robots alongside human workers requires training programs, clear protocols for human-robot collaboration, and cultural change management. Early deployments have found that worker acceptance improves significantly when robots handle ergonomically difficult or dangerous tasks rather than replacing existing jobs.
The humanoid robot market is projected to grow from approximately $2 billion in 2025 to over $30 billion by 2035, driven primarily by manufacturing and logistics deployments.[27] China leads in deployment volume, with companies like UBTECH, Unitree, and Galbot scaling production. The United States leads in per-unit capability, with Figure, Agility Robotics, and Apptronik pushing the boundaries of autonomous task performance.
Key milestones to watch include the first sustained 24-hour autonomous factory operation, the first regulatory approval for humanoid robots in healthcare patient handling, and the first consumer home robot reaching 10,000 units sold.
Longer-range forecasts steepened in 2025. Morgan Stanley projected in April 2025 that the humanoid robot market could reach about $5 trillion by 2050,[51] implying roughly 1 billion humanoids in service by mid-century, with around 13 million units expected in service by 2035, mostly in factories and warehouses.[52]
Private and public capital followed the deployments:
| Date | Company | Milestone |
|---|---|---|
| September 2025 | Figure AI | Series C exceeding $1 billion at a $39 billion post-money valuation, led by Parkway Venture Capital with participation from NVIDIA, Intel Capital, Brookfield Asset Management, Qualcomm Ventures, and others [53] |
| December 2025 | Galbot | Raised over $300 million at a $3 billion valuation, bringing cumulative funding to about $800 million [44] |
| February 2026 | Apptronik | $520 million Series A extension at an approximately $5 billion valuation, lifting total Series A financing above $935 million [54] |
| March 2026 | Unitree | Filed for a STAR Market IPO seeking about 4.2 billion yuan ($610 million); the Shanghai Stock Exchange listing committee approved the application on June 1, 2026 [55][56] |
Unitree's IPO prospectus reported 2025 revenue of 1.71 billion yuan (about $250 million), with more than 5,500 humanoid robots shipped during 2025, a volume the company says ranked first globally, and an average humanoid selling price that fell from 593,400 yuan in 2023 to 167,600 yuan in 2025.[55] UBTECH, meanwhile, reported cumulative Walker-series orders exceeding 800 million yuan by November 2025 and targeted annual production capacity of 5,000 industrial humanoids in 2026.[31]