Physical AI: Quadruped market survey

This is Part 3 of our Physical AI series. Read the other articles: Part 1: Quadruped control architecture · Part 2: Quadruped mechanics.

Quadrupeds are being used for industrial inspection, research, security, and work in places that are difficult or hazardous for people to reach. The robots look similar, but their customers and business models differ considerably. A small platform bought by a university lab is a different product from a robot supplied with inspection sensors, site integration, and a maintenance contract.

This survey looks at the companies serving those markets and the choices behind their products. It also considers India, where research institutions, defence programmes, and domestic manufacturers are beginning to develop a market of their own. The aim is to understand what is being sold, where it is being used, and how much the available evidence tells us about adoption.

Where quadrupeds are useful

The practical advantage of a quadruped is its ability to use separate footholds rather than needing a continuous surface. Stairs, gaps, loose ground, and changes in height can make a route unsuitable for a wheeled robot while leaving a path that a legged robot can follow. Four legs also allow the robot to retain a relatively broad support area when moving slowly or standing still.

This does not make quadrupeds the best option for every task. Wheels are usually more efficient on suitable ground, and a drone may be more useful when the job requires an aerial view. A quadruped becomes relevant when the route requires legs and the task benefits from carrying a sensor or tool close to the ground. Payload, battery life, access restrictions, and maintenance requirements still affect whether it is a sensible choice.

The applications fall into several groups:

• Inspection and mapping. Carrying thermal cameras, acoustic sensors, gas detectors, or laser scanners around facilities, then returning measurements that operators can compare over time.
• Security and hazardous-site access. Providing remote observation in areas where sending a person would create additional risk.
• Research and education. Giving laboratories a physical platform for control, reinforcement learning, perception, and multi-robot experiments.
• Entertainment and broadcasting. Supporting demonstrations, events, and camera work. These uses make the robots familiar to a wider audience, although they say little about industrial reliability.
• Emerging applications. Search and rescue, delivery, and mobile manipulation remain areas of development. Their maturity varies by platform and deployment.

The earlier articles in this series cover the mechanics and control behind these capabilities. For a market comparison, the important distinction is between a robot that can traverse a site and a complete system that can perform useful work there repeatedly.

From research machines to commercial products

Early quadruped work was concerned with whether a legged machine could carry a load over difficult ground at all. General Electric’s Walking Truck, developed under Ralph Mosher during the 1960s, used hydraulic legs controlled by a human operator. The US National Archives documents the programme as part of a broader effort to develop vehicles that could travel beyond roads. Later research programmes, including Shigeo Hirose’s TITAN robots, investigated sensing and programmable walking. US National Archives, Hirose and colleagues’ TITAN III research

Boston Dynamics’ BigDog brought dynamic balance and disturbance recovery to much wider attention in the 2000s. Subsequent machines became smaller and more suitable for operation around people and existing buildings. This development took place alongside work on electric actuation, state estimation, planning, and learned control; it cannot be attributed to a single breakthrough or demonstration. Boston Dynamics history

The commercial transition was gradual. ANYbotics recorded its first sales in 2017 and an offshore deployment in 2018. Boston Dynamics opened commercial sales of Spot in June 2020 after an Early Adopter Program involving more than 150 robots. Unitree was also developing a commercial product range during this period. ETH Zurich, Spot’s commercial launch, Unitree’s product history

These early customers helped establish which capabilities mattered outside a laboratory. Industrial users needed repeatable measurements, dependable navigation, and support when something failed. Researchers needed access to observations and control interfaces. Consumer buyers needed a lower price and a platform they could use without developing their own software.

How to read the market

There is no single measure that captures the quadruped market well. Shipment volume favours lower-cost products, while revenue can be concentrated in a smaller number of industrial systems. A sale may include only the robot, or it may include sensors, software, deployment work, and support. Company-wide figures may also include humanoids, wheeled robots, or components.

Unitree’s disclosures provide one indication of scale. Reporting on its March 2026 IPO filing describes more than 30,000 quadrupeds sold between 2022 and September 2025. That is a cumulative company figure, not a total for the global market. The Shanghai Stock Exchange subsequently announced Unitree’s listing in August 2026. Rest of World, Shanghai Stock Exchange

Annual shipment and market-share comparisons need matching reporting periods and product definitions. Without those, combining figures can give a misleading impression of growth. This survey compares the manufacturers by product positioning and documented use. The table summarises the profiles below rather than ranking technical performance.

Geography helps explain differences in manufacturing costs and customer access, but it does not divide the market neatly. Chinese manufacturers serve different segments from one another, just as Boston Dynamics, ANYbotics, and Ghost Robotics have different commercial priorities.

Boston Dynamics and Spot

Boston Dynamics grew out of the MIT Leg Lab in 1992. Its long development history and public demonstrations gave it substantial visibility before Spot became a commercial product. Hyundai Motor Group acquired an 80 percent stake in June 2021, adding an industrial partner to that research background. Boston Dynamics’ launch announcement, Hyundai’s acquisition announcement

The Early Adopter Program put Spot into construction, power generation, nuclear sites, factories, and research facilities. Boston Dynamics’ launch report describes work with Pomerleau, Hensel Phelps, and Aker BP, among others. This gave the company experience with customers whose requirements went beyond walking: they needed construction records, inspection data, or access to hazardous areas. Early Adopter Program results

An inspection platform with configurable payloads

Spot’s commercial offering is built around carrying equipment and integrating it into an inspection workflow. Its payload range includes thermal, acoustic, gas-detection, and scanning systems. Different customers can therefore use the same base robot for different measurements. Spot payloads

This is relevant when comparing prices. A basic robot and a configured inspection system are not interchangeable purchases. Payloads, computing hardware, charging equipment, software, training, and support affect the cost of deployment. A low entry price on another platform does not establish that it can replace the full system.

Software access

Boston Dynamics publishes SDK documentation, client libraries, examples, and API definitions for developing applications and payloads. It also offers a controlled-access Joint Control API that requires a special-permissions licence. Spot SDK, Joint Control API

For a research buyer, the useful questions are which interfaces are included, which require an additional licence, and whether those interfaces support the intended experiment. For an industrial buyer, the existing inspection tools and integration support may matter more than direct access to the joint controller.

ANYbotics and ANYmal

ANYbotics was founded in 2016 to commercialise quadruped research from ETH Zurich. ETH’s account records first sales in 2017, an offshore deployment in 2018, and the announcement of ANYmal C in 2019. The research group and its early customers provided a route from laboratory development into industrial use. ETH Zurich’s company history

The company has concentrated on inspection in sectors such as energy, chemicals, and metals. Its published customer material includes Outokumpu and Petrobras. These are environments where reducing the need for people to enter hazardous areas can be as important as reducing the time spent collecting measurements. ANYbotics company information

Environmental protection and hazardous areas

ANYmal X has ATEX and IECEx certification up to Zone 1 IIB, as well as IP67 ingress protection. These specifications address different requirements: protection from water and dust does not itself establish suitability for an explosive atmosphere. The relevant certification and configuration need to match the intended site. ANYmal X

This distinction gives ANYbotics a clear industrial focus. Where a site requires explosion-protected equipment, certification can determine whether a robot may be used at all. It is more informative than a general claim that one platform is more rugged than another.

Charging and inspection data

ANYmal can return to a docking station when its battery is low. This supports repeated inspection missions, but it does not mean the robot walks continuously without charging. Mission duration, charging time, route length, and the number of robots all affect coverage. ANYmal’s inspection workflow

ANYbotics also supplies Data Navigator, introduced in February 2025 to organise and analyse inspection data. In our assessment, this software is an important part of the company’s commercial model: once inspection records and routines are integrated into a customer’s operations, changing suppliers involves more than replacing the robot. That can encourage continuity, though data access and integration options should be considered alongside the hardware. Data Navigator announcement

Ghost Robotics and Vision 60

Ghost Robotics has developed a substantial defence and security presence around Vision 60. Its products have been evaluated for tasks such as remote observation and access to difficult terrain. US Department of Homeland Security material documents evaluation work rather than establishing that every proposed application became an operational deployment. DHS Vision 60 evaluation

Government-funded development can shape a product differently from consumer sales. A customer may fund a capability before purchasing the resulting system, while field use places particular demands on repair, transport, communications, and payload integration. These considerations help explain Vision 60’s emphasis on modular hardware.

Operation with limited visibility

Ghost Robotics advertises a blind mode for movement when visual sensing is unavailable or impaired. Its current specifications also list IP67 protection, an operating range of −40°C to 55°C, and field-replaceable legs. The manufacturer states that cold starting is not possible below −20°C, illustrating why a headline operating range needs to be read with its conditions. Vision 60 specifications

The distinction between balance and navigation matters here. A controller may keep a robot moving through proprioceptive feedback while the wider system still needs sensors to identify obstacles, localise, or choose a safe route. Blind-mode movement should not be read as a guarantee of complete autonomy in every environment, or as proof that every competing quadruped stops when a camera is obscured.

Maintenance and integration

Vision 60’s modular construction and open integration architecture are intended to let users adapt payloads and replace assemblies in the field. For customers operating away from a workshop, the time needed to restore a damaged robot may matter as much as its maximum speed. Vision 60 product information

The commercial question is therefore not just whether the robot can complete a demonstration. It is whether operators can maintain it, obtain parts, integrate their equipment, and keep it available over an extended assignment.

Unitree and the Go2 family

Unitree has made quadrupeds accessible to a broader group of buyers through lower-priced products, direct sales, and developer tools. Its product history predates Go2 and includes Laikago, AlienGo, and A1. Go2 continues that approach with configurations aimed at different kinds of users. Unitree’s product history

Consumer demand, demonstrations, and academic use can reinforce one another. More robots in circulation create more examples, software, and people familiar with the platform. For a laboratory choosing hardware, that community can reduce the effort involved in starting an experiment. It does not, however, make every consumer configuration suitable for research.

Price and developer access

Unitree’s official store advertises the Go2 range from US$1,600 and lists the selected Go2 Pro configuration at US$2,800. EDU pricing is provided through sales. Shipping, customs duties, taxes, accessories, and support are additional considerations, particularly for an Indian buyer. Unitree Go2 store

The official comparison distinguishes development access by variant. Air and Pro are not marked as supporting secondary development, X is marked for partial support, and EDU for support. A buyer should confirm the precise interfaces and permissions in the current development documentation rather than infer them from the robot’s appearance. Go2 model comparison

The lower entry price makes experimentation more accessible, but the comparison must use the version that can actually run the experiment. Buying several consumer robots is not a substitute for buying one with the required control access.

Manufacturing and the developer ecosystem

Manufacturing scale and direct distribution help explain how Unitree reaches a wider range of buyers. Its product materials also describe in-house development of components such as lidar. These choices can reduce dependence on external suppliers and give the company more control over product cost. They do not establish a precise manufacturing cost for every model: a company-wide cost average cannot be compared directly with the retail price of its cheapest robot. Go2 product information

Unitree publishes SDK and robotics software repositories. For supported hardware, these give developers a starting point for integration and control work. The practical value comes from the combination of available hardware, usable interfaces, documentation, and examples, rather than from the number of repositories alone. Unitree Robotics on GitHub

The consumer features also serve a different need. App control, follow functions, and built-in demonstrations let buyers use the robot without writing software. Those features broaden access, while research and industrial customers still need to evaluate the interfaces, reliability, and support relevant to their work.

DEEP Robotics and Jueying X30

DEEP Robotics has focused strongly on industrial applications, including power inspection, tunnels, and emergency response. Its X30 is positioned around these tasks rather than as a low-cost consumer robot. The distinction is important: Unitree’s visibility and unit volume do not imply that every Chinese manufacturer follows the same strategy.

Reporting on DEEP Robotics’ IPO application, accepted in May 2026, gives annual revenue of approximately RMB 50 million in 2023, RMB 103 million in 2024, and RMB 337 million in 2025. These figures describe the company’s growth rather than the size of the overall quadruped market. 36Kr’s filing coverage

Industrial specifications

The manufacturer’s X30 page lists a weight of 56 kg including the battery, IP67 protection, an operating range of −20°C to 55°C, slope capability up to 45°, and obstacle clearance of at least 20 cm. Listed battery endurance is 2.5–4 hours, with quick battery replacement. DEEP Robotics explicitly notes that these are laboratory parameters and that real-world performance can differ. X30 specifications

These specifications describe the conditions the platform is designed to handle. They do not establish that it is the fastest or most capable quadruped in every task. Inspection performance also depends on payload quality, navigation reliability, the route, and the customer’s data requirements. Continuous site coverage may be possible through charging, battery changes, or multiple robots; it should not be confused with uninterrupted operation on one battery.

Deployment evidence

Singapore’s SP Group describes using a robotic dog named SPock for independent inspections of power tunnels. DEEP Robotics identifies this platform as an X30. This is useful evidence because the operator describes how the robot fits into its maintenance work, rather than only showing the robot moving through a test environment. SP Group’s account, DEEP Robotics’ deployment report

DEEP Robotics also publishes accounts of substation inspection. Such case studies are useful when they explain the route, equipment inspected, and comparison with the previous process. Vendor-reported savings should still be treated as results for a particular deployment, not as a guaranteed reduction in cost at another site. Substation inspection case study

The Indian market

The Indian market includes domestic robot developers, imported platforms, research purchases, and defence programmes. Public information is uneven, so it is difficult to assign it a precise number of years behind another country. It is also important to distinguish an Indian supplier, an assembled platform, and a robot whose core technology was developed domestically.

Defence is one visible source of demand. Reporting reproduced in DRDO’s September 2024 press digest states that the Army had procured and inducted 100 robotic mules in forward areas. AeroArc also describes its role in supplying robotic mules. These reports establish procurement activity, but they do not by themselves establish a complete replacement timetable for animal transport or quantify the share of indigenous technology in every subsystem. DRDO press digest, AeroArc

The opportunities for domestic companies extend beyond the robot’s purchase price. Local maintenance, spare parts, training, and access to the development team can matter to both laboratories and field users. Establishing those capabilities takes time, and an announcement of a new platform is only an early part of that process.

xTerra Robotics

xTerra follows a research-to-product model. IIT Kanpur describes the company as a startup founded in 2023 at its Startup Incubation and Innovation Centre, developing autonomous legged robots for inspection, data collection, and security. Its SVAN M2 is a quadruped platform, and the company also publishes an M2 SDK with interfaces for developers. IIT Kanpur, xTerra’s M2 SDK

xTerra has also reported delivering an SVAN M2 inspection robot to Siemens Energy, with plans to develop applications for the energy sector together. This provides a concrete example of industrial engagement, although the announcement does not establish the scale of subsequent deployment. Its academic origins and local engineering team also give it a route into research use and platform customisation. xTerra’s delivery announcement

The current company website presents SVAN alongside its wider autonomous-systems work. For a purchasing decision, the delivered robot’s payload, runtime, control access, and support terms should be checked against the current specification. Research deliveries and industrial customers are both relevant signals, but neither should be turned into a claim of large-scale industrial deployment without further evidence. xTerra Robotics

Addverb and Trakr

Addverb enters the category from warehouse automation rather than from a business built solely around quadrupeds. Its existing manufacturing and enterprise experience may help it develop and support a new platform, but the performance of its warehouse business should not be used as a measure of Trakr sales.

Addverb introduced Trakr at LogiMAT India 2024 and publicised Trakr 2.0 at the following year’s event. The 2.0 announcement states a payload capacity of up to 20 kg. Other Trakr material describes configurations with different payload figures, so specifications from the two generations should not be combined into a single comparison. Trakr overview, Trakr 2.0 announcement

The company discusses inspection, surveillance, security, and broadcasting as potential applications, and its material refers to quadruped use during IPL 2025. These uses provide visibility, but the next question for the industrial market is how Trakr performs in sustained customer operations. Named deployments, availability figures, and repeat orders would provide stronger evidence than the size of Addverb’s wider customer list.

Svaya Robotics

Svaya’s quadruped work has been associated with DRDO. The organisation’s March 2023 press digest includes reporting on the Hyderabad company’s collaboration with R&DE in Pune and DEBEL in Bengaluru on a quadruped and an exoskeleton. The quadruped was presented as a technology demonstrator for defence-related use. DRDO press digest

A demonstrator and a product in recurring deployment are different milestones. Public reporting gives less detail about subsequent quadruped orders, operating results, and product revisions. That limits what can be concluded from this survey; it is not evidence that the programme has either stopped or reached commercial scale. Defence-related development may also have reasons for publishing less information than a consumer product company.

Other teams

Strider Robotics and General Autonomy also present quadruped work on their websites. They are worth following as the domestic ecosystem develops, but this survey does not assign them shipment estimates or market positions without comparable disclosures. Strider Robotics, General Autonomy

Across these companies, the useful indicators will be the same: delivered systems, repeat customers, support capacity, and evidence that the robots complete useful work outside demonstrations. The number of teams building a quadruped is only one part of the market’s development.

What may shape adoption next

Our expectation is that near-term adoption will remain concentrated in tasks with a clear operational need, particularly inspection and remote access. A quadruped is easier to justify when it can gather a required measurement, reduce exposure to a hazard, or reach equipment that existing automation cannot access. Its value is harder to establish when the task can already be done reliably by a simpler machine.

The main constraints are likely to include integration and support as well as mobility. A robot needs to connect to the customer’s data systems, complete its route reliably, recover from interruptions, and remain serviceable. An inspection that still needs frequent human intervention may be useful, but its economics differ from a routine that can run with little supervision.

Mixed fleets are a reasonable longer-term possibility. Quadrupeds, wheeled robots, humanoids, and fixed equipment can serve different parts of a workflow. Coordinating them will require shared task information, safety procedures, and reliable hand-offs. The existence of each robot type does not yet establish that the combined workflow is ready for widespread use.

For India, research institutions and defence programmes provide routes into early use, while industrial deployment will depend on proving reliability and service at customer sites. Domestic development can help with adaptation and maintenance, but customers will still need evidence of performance and cost over time.

The manufacturers in this survey are approaching those requirements differently. Some focus on an integrated inspection service, others on field operation or accessible research hardware. Following those differences is more useful than treating quadrupeds as one uniform product category or assuming that a successful demonstration predicts adoption.

References

How to read the sources

Sources are linked beside the claims they support. Company pages establish advertised capabilities and product positioning; customer accounts provide deployment context; financial reporting is identified separately. None of these substitutes for a common, independent benchmark across the robots.

• History: US National Archives, Boston Dynamics, and ETH Zurich.
• Market disclosures: Unitree filing coverage, Unitree listing announcement, and DEEP Robotics filing coverage.
• Global products: Spot SDK, ANYmal, Vision 60, Go2, and X30.
• Indian programmes: IIT Kanpur on xTerra, Addverb’s Trakr 2.0 announcement, and the DRDO press digests for March 2023 and September 2024.