At the beginning of this year, I reached out to the CMO of a leading robotics company, hoping to produce a special feature on the collaborative development of 5G-Advanced and humanoid robots. Unexpectedly, the request was directly declined. In his view, the company's current priorities are twofold: first, focusing on the R&D of the robot hardware and the iteration of large models, where refining the robot's "brain" and "cerebellum" is paramount; second, attracting more attention from the capital market. The synergy between robots and communication networks is not yet considered a critical issue.
However, in just half a year, industry perception has undergone a significant reversal. As humanoid robots transition from laboratories to competition arenas, and further into real-world scenarios such as industrial, household, and logistics applications, an increasing number of practitioners realize that merely perfecting the robot's "brain" and "cerebellum" is far from sufficient. To make robots truly move and come alive, a robust communication network is indispensable to efficiently link massive perception data, control commands, and AI computing information. The recently concluded 2nd World Humanoid Robot Games provided a highly convincing practical validation of this industry proposition.
The scale of this year's World Humanoid Robot Games is unprecedented, featuring over 600 teams and 2,056 humanoid robots competing across 51 events. These include high-intensity confrontations such as table tennis and freestyle combat, dance competitions testing the synchronization of multi-robot movements, and real-world scenario events tailored for industrial, hotel, and logistics applications. With thousands of robots operating simultaneously, they must transmit onboard sensor data in real time, receive millisecond-level motion control commands, and execute multi-robot collaborative operations. Additionally, they need to simultaneously stream ultra-high-definition competition footage, while tens of thousands of on-site spectators access the network concurrently. The traditional network system, primarily designed for human internet access, can no longer withstand the pressure of such a complex "human + robot" mixed scenario.
As the exclusive communication service partner for the event, China Unicom rolled out a comprehensive solution featuring "two networks and one platform": a 5G-Advanced high-uplink network and an all-optical Wi-Fi private network, paired with the country's first embodied AI robot management platform for event scenarios, thereby building an intelligent communication foundation for human-machine symbiosis. Field tests at the venue showed a peak uplink rate of 1Gbps, with average latency kept under 30 milliseconds. By leveraging AI-driven intelligent scheduling algorithms to dynamically allocate network resources, the solution achieves physical isolation between the competition service network and the media network. This prevents signal interference and network contention in high-density scenarios, ensuring stable interaction and precise control for thousands of robots. Meanwhile, the management platform provides a comprehensive overview of location, operational status, and competition progress via dedicated robot network IDs, forming a complete closed-loop operational system. It is this invisible network foundation that enabled over 2,000 robots to successfully complete various high-difficulty competitions.
This practical test has exposed a long-standing cognitive blind spot in the humanoid robot industry. For a long time, the industry's focus has been concentrated on motors, reducers, and large model algorithms—in other words, the robot's physical hardware and AI brain. Network connectivity was generally considered an "add-on," with the assumption that robots could achieve a high degree of local autonomy, making the network merely a nice-to-have feature. However, the reality of thousands of robots competing on the same stage demonstrates that scaling humanoid robots cannot bypass communication networks.
On the one hand, humanoid robots must perform complex environmental perception, where cameras and force sensors continuously generate massive amounts of data. Relying entirely on local processing would lead to practical issues such as high hardware costs, excessive power consumption, and a significant increase in the robot's weight. By leveraging the ultra-high uplink bandwidth of 5G-Advanced, robots can upload perception data to edge computing nodes for large model inference, and then receive control commands back at the robot body. This empowers the terminal robots with cloud-based large model capabilities, realizing "cloud-edge-device" collaboration. Without a communication network offering high bandwidth, low latency, and high reliability, the powerful capabilities of cloud-based large models cannot be seamlessly delivered to physical robots, and so-called embodied intelligence would be confined to standalone local operation.
On the other hand, future humanoid robots will not operate in isolation. Whether it involves collaborative operations among multiple robots in a factory, coordinated interactions among several service robots in a household, or synchronized cluster performances, it requires state synchronization and command coordination across multiple robots. This necessitates a unified "nerve center" network to handle scheduling, state monitoring, and fault alerts. If network latency jitter is high and connections are unstable, robot movements will become disordered, causing collaborative tasks to fail entirely. Take the dance competition at this event, for example: dozens of robots moved in perfect unison, which was made possible by unified command distribution and state feedback guaranteed by a stable network.
Of course, this is not to deny the value of a robot's local autonomous capabilities. Under extreme network disconnection conditions, robots must possess basic local safety and hazard-avoidance capabilities; this is the absolute baseline. However, this does not mean the value of the network can be diminished. Local intelligence and network empowerment are not mutually exclusive but complementary. Local processing handles immediate safety reactions, while the network manages high-computing-power calls, cluster collaboration, remote operation and maintenance, and data iteration. Only by combining the two can the boundaries of a robot's capabilities be continuously expanded.
It is also important to recognize that communication networks designed for humanoid robots operate on a completely different logic than traditional networks built for mobile phone consumers. Mobile services primarily focus on downlink downloads, whereas humanoid robots require the exact opposite, with high uplink capacity being a core necessity. Robots continuously transmit sensor images and motion state data, placing extremely high demands on uplink bandwidth. Millisecond-level control command interactions impose strict standards on latency, jitter, and reliability. Furthermore, the network must support the simultaneous access of massive numbers of robot terminals, enabling terminal identity recognition, state perception, and security management. This compels operators to break away from traditional To C network construction paradigms, build private network systems tailored for embodied AI, customize scheduling algorithms, and develop robot management and operation platforms to achieve a paradigm upgrade in network capabilities. China Unicom's implementation at the competition venue serves as a crucial exploration in this direction.
The competition venue is a testing ground for technology, not the final destination. After the games conclude, humanoid robots will eventually step out of the arena and into factory workshops, hotels, supermarkets, and millions of households. In industrial scenarios, robots must transmit real-time operational data in complex production environments. In household scenarios, they need to connect to cloud-based large models to execute complex tasks. In logistics scenarios, multi-robot cluster scheduling is inseparable from a high-quality communication foundation.
From an industrial development perspective, robotics companies are deeply engaged in physical hardware and AI algorithms, while communication operators are refining 5G-Advanced, all-optical networks, and embodied AI management platforms. These are not two non-intersecting tracks; rather, they must be deeply integrated. If companies solely focus on refining the robot hardware while ignoring network coordination design, they will encounter insurmountable bottlenecks when scaling up for mass deployment.
The concerns raised by that robotics manufacturer half a year ago reflect the prevailing mindset in the industry's early stages. However, following the practical validation at this large-scale event, the industry must re-evaluate the role of networks: communication networks are not merely optional supporting elements for humanoid robots, but the critical foundation underpinning the large-scale deployment of embodied AI. While robots refine their own "brain and cerebellum," communication networks weave the "neural pathways" connecting the cloud, edge, and devices. Through hardware-software synergy and cloud-network integration, the humanoid robot industry can truly achieve steady and sustainable progress.
-END-
Editor/Layout: Zhu Wenfeng
Proofreader: Mei Yaxin
Supervisor: Liu Qicheng