See the real progress of embodied intelligence through the 2025 World Robot Conference (WRC).
This year, humanoid robots remain the core of attention at the event, but the focus of the exhibition has shifted from "being able to walk and move" to "being able to perform tasks and go online". At the exhibition booths, instead of just single-machine demonstrations, real business processes are simulated, with multi-machine collaboration and system integration. Many products directly provide prices, delivery cycles, and maintenance plans.
A few years ago, robots frequently appeared at exhibitions, mainly for showy performances and creating a lot of topics. However, in terms of practical applications, there was still a long way to go before large-scale deployment.
Now, this situation is changing.
In 2025, with the maturity of hardware, the reduction of costs, and the implementation of key technologies such as large models, robots have started to enter more specific scenarios. From factory workshops to rehabilitation centers, from security inspections to retail terminals, some robot products already have the initial delivery capabilities and are operating in real environments. Their roles have also shifted from demonstration prototypes to tools and assistants, and some even have a certain degree of autonomous collaboration ability.
The World Robot Conference provides a clear window for observation.
Compared with 2024, the number of participating enterprises has increased this year, and the exhibited products are more focused on usability and replicability. Industries such as manufacturing, healthcare, and services have become the key areas for implementation. Humanoid robots continue to be the core of attention, but the focus of the exhibition has shifted from "being able to walk and move" to "being able to perform tasks and go online". At the exhibition booths, instead of just single-machine demonstrations, real business processes are simulated, with multi-machine collaboration and system integration. Many products directly provide prices, delivery cycles, and maintenance plans.
Through this window, we can see three clear signals:
First, the technical threshold has been crossed. The combination of multi-modal large models and embodied intelligence has significantly improved the stability of robots in perception, understanding, and execution.
Second, commercialization has started to take shape. The reduction of hardware costs and the increase in the penetration rate of domestic components have made it possible for more products to be SKUized and deployed in batches.
Third, the traction of scenarios has been enhanced. High-frequency and controllable tasks in manufacturing, rehabilitation, park security, retail, etc. have started to be piloted in batches, forming a closed-loop operation with the back-end systems.
If in 2024 we saw what robots could do, then in 2025 we will really start to discuss where and how robots can be used.
I. Robots Entering Scenarios
The biggest change presented at the 2025 World Robot Conference is the shift in the demonstration logic.
Compared with 2024, when the emphasis was on what the products could do, more participating enterprises this year are answering the question of "whether they can be used". The booth design has also shifted from single-machine demonstrations to multi-robot collaboration, toolchain displays, and restoration of typical scenarios, with the focus on whether the products have deployment capabilities, operational stability, and system integration.
Among them, humanoid robots still occupy the central position, but their roles have obviously changed.
In 2024, humanoid robots mainly had dynamic demonstrations, emphasizing basic movements such as bipedal walking and waving. This year, the humanoid robots brought by enterprises such as Ubtech, Galaxy Universal, Fourier, and Zhipu were demonstrated for the first time in a near-real "work scenario" to perform tasks such as handling, picking, collaborative operations, and even completing process tasks. The Walker S2 demonstrated by Ubtech achieved loading, unloading, and handling operations in a simulated production line environment, while Galaxy Universal demonstrated a complete closed-loop operation of a robot replacing humans in a retail store scenario. Many models are marked with continuous operation time, load indicators, and training methods, and the product form has started to shift from "prototypes" to "SKUized" products.
Traditional industrial robot manufacturers are also adjusting their strategies and actively moving towards scenario integration.
Companies such as JAKA, EVAT, and AUBO, which originally focused on collaborative robotic arms, have concentrated on displaying product lines such as mobile operation platforms (robotic arms + chassis), quick-change modules, and reusable task packages that are more in line with the needs of flexible manufacturing.
Such products usually come with MES/WMS docking solutions, emphasizing the online deployment cycle and the openness of the toolchain. This marks the transformation of traditional industrial robots into the role of "system integration providers".
The changes in the service robot exhibition area reflect that the industry has become more focused after the contraction.
In 2024, there were still a large number of robots for hotel delivery, front desk reception, and display and promotion, and the business models of some products were unclear. This year, most enterprises have concentrated their products on high-frequency, stable, and sustainable scenarios, such as convenience retail, business reception, and indoor delivery. The "Galaxy Space Capsule" project in Beijing is a representative example. In the intelligent retail terminals operating in Haidian District, the robots not only undertake product distribution but also support collaboration with the back-end system to complete replenishment and settlement, forming a mini-store model with a continuous commercial closed-loop.
Medical and rehabilitation robots show a trend of systematization and platformization.
Stardust Intelligence launched three immersive experience and display areas with its Astribot S1. It is reported that this robot was deployed in a nursing home in the first half of the year.
Fourier set up a rehabilitation workstation in the exhibition area, where multiple robots collaborate to support different functional trainings, centrally controlled by a therapist. This solution has been put into use in some top-tier hospitals. Compared with the "single-machine + schematic diagram" display in 2024, this year, the deployment structure, data feedback mechanism, and human-machine collaboration process are clearly shown, with a greater emphasis on the improvement goals of efficiency and the bed output ratio.
Running through these changes is the comprehensive improvement of toolchain capabilities.
Many participating enterprises have brought their in-house development platforms, simulation training systems, and remote operation and maintenance backends to the forefront for display, emphasizing reusable task scripts, online training, remote upgrade, and deployment management capabilities. These invisible basic capabilities are the key to whether robots can be truly implemented, and they were also relatively lacking in the 2024 exhibition.
Overall, the focus of the 2025 Robot Conference has shifted from being able to move to being usable, and is gradually moving towards being able to go online, be maintained, and be iterated. Enterprises are no longer just showing functions, but are starting to talk about pricing, delivery dates, deployment cycles, and maintenance mechanisms. What this reflects is the real progress of robots from technology demonstrations to industrial applications, gradually approaching the "positive ROI".
II. The Driving Forces Behind the Implementation of Robots: From "Showy Performances" to "On-the-Job"
The transformation of robots from being able to perform on the exhibition stage to being usable in real scenarios is not an overnight technological miracle, but the result of the steady evolution and gradual alignment of multiple technological routes, industrial chain links, and market conditions in the past few years.
First and foremost, it is the sinking of large model capabilities and the actual implementation of embodied intelligence.
In the past, robots often relied on rule programming to perform tasks and had difficulty coping with slightly more complex environmental variables. Now, with the addition of multi-modal models, robots can understand human intentions through "seeing" and "hearing".
For example, at the exhibition, some humanoid robots can already receive fuzzy semantic instructions and, combined with visual recognition, spatial perception, and task planning, autonomously complete tasks such as grasping, handling, and even path selection. Although there is still a gap from full generalization, the path from the laboratory to prototype demonstration and then to partial pilot operation has been initially opened up.
Meanwhile, the progress in hardware and manufacturing has solved another obstacle: cost.
It should be noted that a few years ago, the price of a relatively fully functional robot could be as high as one million yuan, and many enterprises were even reluctant to quote prices. Now, at this year's conference, more and more manufacturers are actively marking the prices and configuration details of the whole machines, ranging from one million yuan to several hundred thousand yuan, and some lightweight platforms are even priced below 100,000 yuan.
This is the result of the improvement of the supply capacity of the entire industrial chain. The maturity of domestic components, lightweight modules, and integrated design has greatly improved the manufacturability and delivery capabilities of robot products.
For whole - machine enterprises, this means that products can finally be defined and packaged in the form of SKUs, which also means that robots truly have the preconditions for being circulated as commodities.
If technological progress and cost control have prepared the body for robots, then the clear traction from the scenario side is what makes them really move.
In the past, the biggest problem the robot industry faced was not the inability to produce, but the inability to use. That is, many products could not find suitable application scenarios, or even if they did, it was difficult to close the commercial value loop. However, this situation is changing, and some high-frequency local tasks within the industry are starting to emerge.
For example, in manufacturing, there are tasks such as station handling; in logistics, there is sorting; in rehabilitation medicine, there are standard training processes; in park security, there are night inspections; and in service retail, there is product delivery. These tasks are characterized by repetition, stable rhythm, and controlled space, which are exactly suitable for robots to enter and replace some human labor.
At the conference site, the humanoid or mobile robots exhibited by many enterprises are no longer just for "showing off", but are performing these seemingly ordinary yet real detailed tasks. These are the starting points for robots to truly move towards practical value.
Greater structural support comes from the maturity of policy and ecosystem support.
Compared with 2024, the implementation mechanism of "scenario + policy + enterprise linkage" is clearer this year. Many places have started to promote the trial use of robots locally, no longer only focusing on technological debuts or industrial layouts, but clearly specifying task indicators, operation cycles, and support methods.
For example, some parks provide trial operation space and data access interfaces for robots; some hospitals cooperate with rehabilitation equipment for actual bed scheduling; and some cities have started testing new models such as "robot + convenience retail" or "robot + intelligent transportation". This means that the threshold for commercial verification is decreasing, and the growth path is becoming more predictable.
Looking back at these series of changes, it can be said that the robot industry is stepping in the right rhythm from multiple directions simultaneously. No single step is a decisive breakthrough, but all variables have been aligned this year, making it possible for robots to move from demonstration to practical use.
III. Easy to Enter the Stage, Difficult to Stand Firm: The Next Challenge for Robot Implementation
Although the 2025 Robot Conference gives hope for implementation, the real challenges are just beginning. In the next few years, the development of the industry will mainly proceed in two directions: the depth and breadth of scenario penetration, and the maturity of business models and operation systems.
In terms of scenario penetration, robots will move from single-machine operations to system collaboration.
Among them, the manufacturing and logistics industries may be the first to complete this transformation. In future factories, humanoid robots and mobile operation platforms will seamlessly collaborate with robotic arms and automatic conveyor lines to complete material transfer across workstations and regions and flexible section operations. The robot fleets in logistics parks will achieve automatic scheduling, and the entire process from sorting to loading and unloading will be unmanned.
In the service field, the deployment density of robots in high-frequency scenarios such as retail, catering, and security will continue to increase, expanding from isolated pilot points to an organized and scheduled operation network. By then, robots in the city may become part of the infrastructure, just like express delivery cabinets today.
The evolution of business models will be synchronized with this. The proportion of one-time equipment sales will decrease, and Robot as a Service (RaaS) will become the mainstream.
That is, customers pay fees monthly or annually to obtain a complete service package including equipment, maintenance, upgrades, and insurance. Manufacturers monitor the equipment status through a remote operation and maintenance platform, push task updates, and optimize performance.
This model not only reduces the initial investment of customers but also allows manufacturers to continuously accumulate operation data in long-term services, forming a two-way cycle of technology and business. With the accumulation of operation data, robot operators may form economies of scale and technological moats, just like today's cloud computing service providers.
However, an optimistic trend does not mean fewer challenges.
Firstly, there is the issue of reliability. Many robots perform excellently in exhibition and pilot environments, but in real long-term operation scenarios, they are still affected by environmental factors such as light, dust, and temperature differences, resulting in perception deviations or unstable movements. The stability of key technologies such as dexterous hands and force control modules still needs to be improved when dealing with fine operations on flexible objects and complex surfaces.
Energy efficiency is also a limiting factor. Especially in large mobile platforms and humanoid robots, the balance between computing power requirements and battery life has not been fully resolved.
Secondly, there is the cost of system integration. Although the hardware price has decreased, seamlessly integrating robots into existing information systems such as WMS, MES, and security platforms still requires high customized investment, which greatly affects the deployment speed and the ability of large-scale replication. Many enterprises find that the online deployment cycle is more than twice as long as expected when introducing robots, which causes some projects to stagnate after the pilot phase.
It can be predicted that the next three to five years will be a critical stage for the robot industry.
On the one hand, the simultaneous maturity of technology, cost, scenarios, and policies has created a historical opportunity for large-scale deployment. On the other hand, the challenges of stability, integration, and social adaptability determine that this process will not be smooth sailing.
Looking at the current situation, those who can accumulate experience in real operation and establish replicable implementation paths will take the initiative in the next round of competition.