The robot competition arena has begun to weed out "one-trick ponies"

8.64 seconds for 100 meters — the moment when the humanoid robot Tiangong Ultra refreshed the human record is one of the most exciting highlights of this World Humanoid Robot Games. Its "small-sized" teammate Tiangong Omni claimed the championship of the 400m small-sized group with a time of 45.66 seconds, and its slightly comical "face-covering running" pose quickly went viral across social media.

On the other side of the track, the atmosphere was far quieter. In the scenario competitions, robots were assigned to complete various tasks covering new energy vehicle charging, book sorting, hotel services, industrial operation, garden inspection and emergency rescue. These events rarely have the dramatic scenes when a new record is set, but they are much closer to the scenarios where people are willing to pay for robots.

The two types of competitions just present two major challenges for the humanoid robot industry in the same venue: one side pushes the physical limit, and the other side tests the completion of real tasks.

Beijing Humanoid Robotics Innovation Center Co., Ltd. (hereinafter referred to as "Beijing Humanoid") has delivered an impressive result. It participated in 41 events with four robots including Tiangong Ultra, Tiangong Omni, Tiangong 3.0 and Tianyi 2.5-TE, and finally won 15 gold medals, 12 silver medals and 18 bronze medals, ranking first in the total number of medals (data from official announcements). It stood out in both the athletic competitions and scenario competitions, and is one of the all-round players rarely seen in this event.

What is more noteworthy is that this report card, which looks very much like a commercial company showing its strength, comes from a national-local co-built innovation center. Commercial companies usually prove their strength through orders, while Beijing Humanoid, while focusing on immediate demands, also heavily invests in technologies whose market returns are not yet clear. Now, this platform route that requires heavier investment and has a longer cycle is ushering in its payback period.

01. The Robot Arena Eliminates "Partial Subject Underachievers"

In the past few years, humanoid robots have been particularly good at creating hot topics, and a 10-second video of boxing or dancing can attract massive attention. But at the games, they faced a completely different environment: no retakes, no post-editing. Under live broadcast, countless mobile phone lenses and spotlights, the arena exposes the real capabilities of robots to timers and task rules.

When it comes to excitement, track and field events definitely rank high on the list. The 100m tests explosive power, the 400m starts to challenge curve control, and the 1500m requires a balance between speed and energy consumption. Tiangong Ultra not only set the new 100m record of 8.64 seconds, but also won the 400m gold medal with 38.15 seconds, and claimed the 1500m championship with 2 minutes 21.64 seconds. Once the speed of the robot increases, it poses huge challenges to the joints, motors and control systems.

The faster the speed, the greater the landing impact and joint load the robot bears, and the higher the motor temperature will be. This requires the R&D team to find a balance between body weight, power and heat dissipation, and the algorithm must correct the gait at any time. Any broken link may lead to a loss of the game at any time.

The rules of scenario competitions are more complex. Taking the "smart charging service post" as an example, the robot needs to open the charging port covers of three different vehicles, and then complete the actions of picking up, inserting and pulling out the charging gun with accurate movements. The book sorting event is similar: the robot needs to learn to judge the position of books and bookshelves so as to "pick up" and "place" correctly. Even if the surrounding environment changes, the robot can still complete the task, which is the embodiment of generalization capability.

At the closing ceremony, the "Comprehensive Five Events" as the finale of the scenario competitions is a more complete display of Beijing Humanoid's technical philosophy. Tiangong 3.0 won both the gold and silver medals with its whole-body control (WBC) technology, and became the only biped humanoid robot that won a gold medal in the scenario competitions. Unlike most players in the industry who apply whole-body control to demonstration movements such as dancing and gymnastics, Beijing Humanoid further promotes WBC to real task execution, which requires the robot not only to "complete the movements", but also to perceive and operate while moving, coordinate the legs, torso and arms in a unified manner, and complete whole-body movements such as grasping, handling and interaction during movement.

This requires simultaneous processing of complex problems such as dynamic balance, multi-contact constraints, hand-foot coordination and external disturbances, putting forward higher requirements for control accuracy, real-time performance and robustness.

If the track and field arena pushes the instantaneous performance to the limit, scenario tasks test the reliability and generalization capability of robots. Both capabilities are indispensable puzzles that general humanoid robots must complete. Especially in low-fault-tolerance scenarios such as factories and rescue, one mistake may interrupt the whole process, or even lead to safety risks.

It is not easy to balance these two capabilities. The body needs to respond quickly to model instructions, and feed back the deviation during execution to the system in real time. The faster the speed, the shorter the time left for error correction; the longer the task, the higher the probability of error accumulation. Algorithms and hardware need to be repeatedly adjusted and polished through a large number of practical experiences, which is why embodied intelligence is more difficult than pure software models.

In this event, Beijing Humanoid sent four "participants" with clear divisions of labor. Tiangong Ultra is responsible for sprint, long-distance running and jumping, pushing the high dynamic motion capability to the upper limit; Tiangong Omni with smaller size is responsible for sprint and obstacle running, verifying perception, movement and agile control; Tiangong 3.0 is responsible for the Comprehensive Five Events, challenging cheerleading, weightlifting, tug-of-war and emergency tasks, and won the championship of the Comprehensive Five Events with obvious advantages; Tianyi 2.5-TE is responsible for operation tasks such as new energy vehicle charging and book sorting, demonstrating long-process operation capability.

In the end, they not only ranked first in the total number of medals, but also became the rare company that fought on two fronts and won victories on both the athletic and scenario tracks in this event.

02. How Is This All-Round Team Built

Among a large number of embodied intelligence manufacturers, why is a national-level innovation center the one that ranks first on the medal table? The answer lies in its platform route.

In November 2023, Jingcheng Electromechanical, UBTECH, Xiaomi Robotics, Yizhuang Robotics and other units jointly initiated the establishment of Beijing Humanoid. At the beginning of its establishment, it listed the general ontology and embodied intelligence model as the key direction for common technology research. In April 2024, the robot mother platform "Embodied Tiangong" was released; in March of the next year, the general embodied intelligence platform "Huis Kaiwu" was launched.

In the same year, Beijing Humanoid was upgraded to the National-Local Co-built Embodied Intelligent Robot Innovation Center, and the parallel development of software and hardware has become the core of its platform route.

"Embodied Tiangong" is a scalable ontology foundation, where the joint system, mechanical configuration and motion capability are precipitated, and then form products such as Tiangong Ultra and Tiangong Omni. "Huis Kaiwu" adopts the architecture of "embodied brain for decision-making, embodied cerebellum for execution": the brain is responsible for parsing instructions and on-site information, and the cerebellum converts decisions into movements.

Both the athletic tasks and scenario tasks in this event are built on these two layers of platforms. The four robots operate separately, but share the same set of technical logic behind them: One brain for multiple robots, one technology for multiple uses — this effectively solves the two most practical costs of robots: development time and delivery cost.

"One brain for multiple robots" refers to the cross-ontology reuse of robot models. Traditional robot systems are usually developed around a specific ontology. After replacing the ontology, the "brain" has to be redeveloped from scratch. The unified embodied model Pelican-Unify released by Beijing Humanoid this year solves this problem: when the same plugging task is transferred to another robot, there is no need to train from scratch. This can not only reduce redundant R&D, but also allow one set of model and data to serve more products.

"One technology for multiple uses" enables robots to realize cross-scenario migration. "Huis Kaiwu" splits a complex task into basic skills such as moving, grasping and placing, and recombines them according to the task requirements. For example, the grasping skill learned by the robot during handling can be used in sorting events; the mature navigation capability can also be applied to charging or inspection tasks.

This idea has achieved obvious effects in practical tasks.

Li Guozhen, head of the embodied intelligence industry application team at Beijing Humanoid Robotics Innovation Center, mentioned when reviewing the scenario competitions that there are many engineering challenges in the new energy vehicle charging scenario. For example, outdoor sunlight will amplify perception errors, the robot may encounter cable winding during movement, and the robot needs to coordinate navigation, obstacle avoidance, motion planning, and finally complete the fine plugging operation.

Thanks to "Huis Kaiwu", the team assigned perception, movement and plugging to different models for processing respectively. Li Guozhen said that even if a problem occurs in one link, the team can optimize that specific link separately, thus reducing the debugging complexity of long-process tasks. At the same time, this architecture can also precipitate capabilities into skill packages, which can be used in unmanned vehicle station energy supplement, highway charging services and other scenarios in the future.

The ontology solves the problem of "what you have", the model solves the problem of "how to use it", and the ecosystem determines "how far you can go". It brings Beijing Humanoid's platform beyond the original R&D team, allowing more parties to participate.

At this games, Beijing Humanoid cooperated with more than 30 universities including Tsinghua University, Peking University and Shenzhen University, as well as industrial partners, to organize more than 60 teams and participate in the competition with nearly 200 robots. Beijing Humanoid opened its platform and toolkits to participants, allowing different teams to develop on the same technical foundation, which also enabled the platform to get more polishing in real scenarios.

The ecosystem is also the only way for embodied intelligence to cross the large-scale application threshold. The industrial tasks faced by robots are highly scattered, and it is difficult for one single company to fully master all processes. After opening the underlying capabilities, industrial partners can solve the problems they are more familiar with, and eventually make the platform capabilities stronger.

Only when technical capabilities go beyond the boundary of one single company, can the experience of all participants be transformed into the common accumulation of the whole industry. This route is destined to take a longer cycle, and it also brings Beijing Humanoid to the next challenge: as a national-level innovation platform, it must also tackle the hard nuts that are difficult to get returns in the short term but determine the future of the industry.

03. The "National Team" Tackles Hard Nuts

The embodied intelligence industry is facing a reality: when more and more robots enter factories and service places, "partial-capability" robots are increasingly difficult to win new orders. Nowadays, robots that can only dance or box are no longer novel. Only systematic capabilities can determine whether an enterprise can cross the most important commercialization threshold.

That is why the "national team" attribute of Beijing Humanoid is particularly important. It not only needs to develop models and complete robots, but also build a foundation for common industrial problems, and promote the flow of achievements to the whole industry.

This means it has to do some work that is not commercially attractive. The R&D cycle of models is very long, and it is difficult for standardized interfaces to generate revenue immediately. Facing survival pressure, pure commercial companies usually prioritize the scenarios that can generate orders faster, while national-level innovation platforms need not only focus on the uncharted technical territory, but also deal with all kinds of "dirty work" in engineering.

Beijing Humanoid chooses to precipitate these underlying investments into an open platform. It opens up both ontology and software: the mechanical drawings, robot description files and SDK of Tiangong have been opened to developers, and "Huis Kaiwu" provides model capabilities and tool chains. Developers can continue to train and deploy on the existing achievements, reducing the cost of repeatedly building basic systems.

The industry often says that there is no need to reinvent the wheel, but for humanoid robots, even the size and interface of the "wheel" have not been unified yet. A large number of adaptation work is required when different ontologies connect to the model, and application parties need to re-understand the underlying system after purchasing robots. The role of Beijing Humanoid lies in "empowerment": it helps partners skip the stage of repeatedly building underlying systems, directly carry out adaptation around industrial demands, and make robots move faster to project delivery.

In other words, this "national team" is taking a more practical industrial route: while promoting technical implementation, it expands scenarios together with partners through the open platform, bringing capabilities to more industries.

As mentioned in the Comprehensive Five Events, Beijing Humanoid is promoting the evolution of WBC from a "single skill controller" to a general whole-body control model. Aiming at the huge difference in motion capability requirements of different tasks in the event, by introducing the MoE multi-model fusion architecture, it unifies different motion and operation capabilities into one set of control framework, dynamically invokes and combines different control capabilities according to robot status, task objectives and environmental changes, and realizes the natural connection of skills such as walking and running, posture and height adjustment, and whole-body coordinated operation. The gold medal won in this event verifies not just a single movement, but the general motion intelligence and whole-body autonomous operation capability of the robot when facing continuous, diverse and unstructured tasks, which is also an important verification for whole-body control technology to move from "high dynamic demonstration" to "working in real scenarios".

For Beijing Humanoid, this is a kind of long-termism: continuously invest in a set of infrastructure that is difficult to monetize in the short term but can save a lot of trial and error costs for the industry. The capabilities it accumulates can therefore go beyond the original R&D team and become the starting point for more robots to move forward