In 2026, the real proving ground for humanoid robots may be narrower than the demos suggest: specific roles with clear responsibilities, measurable outcomes, and customers willing to pay.
The embodied intelligence sector has drawn heavy investor attention this year. Financing announcements have filled industry feeds, while bipedal, wheeled, and biomimetic robots have appeared on event stages kicking soccer balls, dancing, and performing other polished demonstrations.
Behind those displays is a more difficult question: how many of these robots can do useful work at scale? Many companies are still focused on making robots look and move like humans. The commercial path, one that supports scaled delivery, produces clear returns, and keeps robots reliably on the job, remains less certain.
Keenon Robotics, which has spent 16 years in embodied intelligence, is offering its answer.
On May 25, Keenon’s compact humanoid robot, Xman-L1, debuted at a streetwear brand event. The launch filled a gap in its humanoid lineup: full-size humanoid robots carry core service capabilities, while compact humanoid robots add lightweight interaction and motion performance.
The move wasn’t just a response to industry hype. Keenon has shipped more than 100,000 units globally and tested its products across mature commercial settings. Its humanoid strategy is based on what it sees as real scenarios, real customers, and real demand.
“Customers don’t actually have a need for humanoid robots. They only care what you can help them do and how much it costs. They don’t care whether you are humanoid or a box,” said Li Tong, CEO of Keenon Robotics, summing up the current reality. “For us, the goal is to use the most effective form available today to solve the pain points customers may not be able to articulate, but still experience. Humanoid robots just happen to be a new option in the toolbox.”
Keenon no longer wants to be seen only as a service robot company. It is trying to reframe itself as a broader robotics company, with commercial service robots as its first successful use case. Specialized robots provide the base for scaled revenue and customer relationships, while humanoid robots extend its capabilities and complete its product matrix.
Start with demand
While many humanoid robot makers are still debating form, movement, and technical concepts, Keenon has shifted its focus back to a basic industrial question: why are robots being built in the first place?
For many startups, the sequence may appear straightforward: achieve key technological breakthroughs, build a humanoid prototype with a complete form and smooth movement, then look for suitable scenarios and test commercial value in the market.
That technology-first route has potential, but it also carries uncertainty. When a product is not tied to real demand from the start, companies can end up solving problems customers do not have.
Keenon’s path, shaped by 16 years of industry practice and more than 100,000 deployed units, runs in the opposite direction. It starts with customer pain points, uses technology to respond to them, and then chooses the robotic form best suited to the task. For Keenon, robots are not lab-born R&D projects. They are tools for specific roles, scenarios, and customer problems.
That difference is central to how Keenon distinguishes itself from other startups.
Its long presence in high-density service settings, including hotels, restaurants, supermarkets, and hospitals, has shown it a gap that specialized robots cannot fully address. Specialized machines are already capable in delivery, cleaning, and transport, but they cannot cover every service role.
Delivery, cleaning, and transport robots remain Keenon’s core business. These machines are optimized for high-frequency, repetitive, and standardized workflows. They are efficient, cost-controllable, and relatively easy to assess through return-on-investment calculations. Keenon says they have generated repeat purchases from many chain customers worldwide, becoming part of the infrastructure for service automation.
But real service environments are messier than standardized workflows suggest. In a hotel lobby, guests may ask for directions, request luggage storage, or file a sudden complaint. In a restaurant, customers may wave for extra utensils, ask about dishes, or complain about serving order. In a supermarket, shoppers may struggle to find products, need guidance, or encounter an unexpected issue.
These roles are flexible, interactive, cross-task, and nonstandardized. They require mobility, manipulation, and interaction at the same time. They also require robots to handle shifting foot traffic, ad hoc requests, complex communication, and transitions between scenarios. These are the blind spots of specialized robots.
Li told 36Kr that these gaps are among the most difficult for customers to address: “What many roles need is a combination of capabilities. Specialized robots can cover the extreme efficiency of a single function, but they cannot fill these gaps. Those gaps can only continue to be shouldered by people. And labor is becoming more expensive, while the gap is getting bigger.”
Rising labor costs, high turnover, and difficulty standardizing service have long challenged the service industry. Specialized robots have solved part of the problem, but not the underlying need for flexible labor.
Through long-term discussions with global customers, Li said Keenon has found that buyers care about four basic questions: Can the robot take on the job? Will it operate reliably? Is it safe in the intended use case? Is the overall investment worthwhile?
What the robot looks like matters less.
Li’s view is that the market needs robots that solve problems, run reliably, and make economic sense, not conceptual devices packaged around a humanoid form.
The spread of large models and embodied intelligence technologies has improved robots’ environmental perception, task generalization, and multimodal interaction. General-purpose robots are becoming more deployable, at least in constrained settings.
But technical feasibility is not the same as commercial feasibility. As many startups pursue the long-term goal of an all-purpose home robot, Keenon has chosen a narrower entry point. Fully capable general-purpose robots may be the long-term direction, but the development cycle is long and commercialization remains difficult. By comparison, achieving role-level capability offers a more realistic path to nearer-term returns.
“It can complete describable, quantifiable, and verifiable tasks within clear role boundaries, quickly generate value, and accumulate data and iterate capabilities in the process,” Li said. In his view, technological progress, urgent customer demand, and controlled deployment scenarios have converged, making this the right time for Keenon to move into humanoid robots.
Roles come before generalization
Instead of pursuing unlimited generalization in the short term, a robot can first become a reliable “employee” for a single role. It can handle defined tasks, stay within clear boundaries, and deliver steady value. This approach may draw less attention than stage demonstrations, but it is one of the few embodied intelligence application paths now being tested through scaled deployment and repeatable use cases.
Keenon defines role-based deployment as placing robots in real commercial roles where responsibilities can be described, tasks can be quantified, and outcomes can be verified.
The challenge is that service scenarios include both flexible tasks, such as reception, information queries, item delivery, and guest-flow interaction, and standardized roles, such as delivery, cleaning, and transport. The former requires cross-task interaction. The latter depends on efficiency and clear returns.
Using specialized robots to cover every role would raise R&D costs, lengthen development cycles, and make scaling difficult. Expecting general-purpose humanoid robots to do everything would also be inefficient, especially in standardized tasks where simpler machines can perform better at lower cost.
This structural difference means future service scenarios are unlikely to be solved by a single robot form. Keenon’s answer is a division-of-labor system in which specialized robots and general-purpose humanoid robots work together. The model is based on the cost structures, capability boundaries, and commercial returns of each category.
Specialized robots follow an optimization model. A delivery robot repeatedly handles a fixed route from point A to point B. Motor power, sensor layout, and charging strategy are designed around that task. It has few redundant parts and limited excess computing power, so the marginal cost of each task can be low.
General-purpose humanoid robots follow an amortization model. Their R&D and hardware costs are higher, but they can be reused across roles and complete longer task loops. A single humanoid robot could greet guests at a front desk in the morning, guide diners at noon, run interactive displays at a shopping mall in the afternoon, and handle simple item delivery at night. Although a general-purpose robot may be less efficient than a specialized robot in any one task, its overall return may improve when broader role coverage and all-day utilization are considered.
In other words, specialized robots are suited to pushing one task toward maximum efficiency, while general-purpose robots are suited to performing multiple tasks above a baseline standard. Combining the two may offer a better balance between cost and coverage. Each robot performs its own function, and multiple robots collaborate to form a complete service loop.
If collaboration between specialized and general-purpose robots answers which machine does which job, the next question is what products should carry out those roles.
Keenon launched Xman-F1, a full-size bipedal humanoid robot, in July 2025. It targets closed-loop service across complex scenarios such as restaurants, hotels, and retail, and can handle integrated tasks such as greeting, guiding, and item delivery. On May 25, Keenon expanded the lineup with its compact humanoid robot.
With that, Keenon’s embodied robotics portfolio has become clearer:
- Specialized robots will continue to support standardized tasks such as delivery and cleaning, providing the cash flow behind scaled revenue.
- Full-size humanoid robots target core commercial roles in hotels, supermarkets, restaurants, and similar settings, with capabilities in mobility, manipulation, interaction, and generalization.
- Compact humanoid robots emphasize lighter weight, greater flexibility, and stronger motion performance, targeting segmented scenarios such as greeting, interaction, displays, and lightweight performances.
Together, the three categories form a tiered lineup designed to match a wider range of role-based needs, from standardized to flexible tasks, and from high-frequency work to long-tail demand.
Keenon is also pragmatic about mobility systems. While much of the industry is investing in bipedal form factors and stage demonstrations, Li said Keenon currently sees wheeled models as the main direction for commercial deployment. That decision, he added, reflects the company’s experience operating more than 100,000 devices in real-world service environments, where dense foot traffic, complex layouts, and safety requirements shape deployment choices.
“When a bipedal humanoid robot loses power, it can fall uncontrollably, creating major risks in crowded service scenarios. If a robot falls in a hotel lobby, it is not just a question of device damage. It could injure an elderly person or a child, and that is something no commercial company can bear. A wheeled version, by contrast, can remain stable and stationary after losing power and will not topple. In terms of safety, stability, deployment speed, and cost control, it is better suited to large-scale commercial use,” Li told 36Kr.
The core of role-based deployment is stability, safety, and deliverability. At this stage, Keenon sees those as the key advantages of wheeled humanoid robots.
Role-based deployment, collaboration between specialized and general-purpose robots, a layered product matrix, and a wheeled-first approach form the core of Keenon’s embodied intelligence strategy. But a clear strategy is only the beginning. To achieve scaled deployment, companies must compete on more than technical specifications and stage performance.
Commercialization is the test
Technical parameters and lifelike movement may attract attention, but they do not automatically create long-term differentiation. In robotics, the more durable advantage may be end-to-end commercialization: the ability to deploy, mass produce, support, and profit from machines in real environments.
Over 16 years, Keenon has used commercial deployments to build what it sees as comprehensive industrialization capability. That includes scenario adaptation, engineering-led mass production, global delivery, cost control, and customer trust. These capabilities can determine whether a robotics company can move through industry cycles.
The first barrier is scenarios and data. Scenario data is the foundation of technological iteration.
Keenon said it has placed more than 100,000 units into real roles worldwide, accumulating long-term operational data on obstacle avoidance, interaction, and practical use. Because this field data is tied to actual roles, it can support algorithm iteration, improve device stability, shorten the path from prototype to scaled deployment, and create a scenario-data advantage that is difficult for newer entrants to replicate.
Data shapes the product ceiling. Engineering-led mass production sets the deployment floor. The industry’s common pain point is that prototypes are easier to build than stable volume production. Short-term R&D can produce a strong prototype, but supporting millions of stable operations and standardized global delivery requires long-term engineering accumulation. Capital can accelerate this process, but it cannot replace it.
Keenon’s approach is to build its own R&D, production, and supply chain capabilities. From core components to whole-machine integration, and from reliability optimization to cost control, the company says its systems have been tested through more than a decade of engineering work and the real-world validation of more than 100,000 shipments.
Keenon’s overseas work, global footprint, and customer relationships have also created a time-based barrier. The company says its business covers more than 70 countries. Through its channels and after-sales systems, it can use field experience and customer feedback to iterate products. This gives its humanoid robots a platform for proof-of-concept validation and scaled deployment, with advantages in deployment efficiency and tolerance for error.
Across these efforts, Keenon’s guiding idea is that robotics must be anchored in deployment value. Product definition and technology iteration revolve around reliability, safety, and commercial returns. The company is not trying to build conceptual props. It is trying to deliver productivity tools that can take on real roles, run stably, and offer strong cost performance.
Industry trends point in a similar direction. Some humanoid robot makers have begun expanding into nonhumanoid product lines to close commercialization gaps and build a healthier cycle through mature deployment scenarios.
In 2026, the humanoid robot market is moving from concept-driven enthusiasm toward a more demanding phase of commercialization. Mass production does not equal commercialization. Shipments do not equal deployed value. Advanced form does not equal customer recognition. The market’s core demand remains unchanged: machines that solve real problems, replace repetitive work, and generate returns.
Keenon has already validated a development path for embodied intelligence. It is not confined to a single scenario, held hostage by product form, or trapped in between capital-backed factions. Instead, its strategy is to anchor itself in customers’ real needs, focus on commercial deployment, and dig deep into role-based value.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.