From the 400-meter track to the real world, robots must prove their reliability. Robots may cross the finish line and set new records, or they may fall on the curves, or even fail to complete the race due to component issues. At the second World Humanoid Robot Games, these "unexpected events" on the field hold real value in the eyes of Kang Yu, General Manager of the Capital Markets Department at Shoucheng Holdings. "The Games can be understood as a stress test, or a reliability test conducted under the spotlight," Kang Yu told reporters, including Yicai, during the event. Shoucheng Holdings is one of the deep participants in the World Humanoid Robot Games, engaging in robot equipment, training and testing, industrial services, and scenario support through its robotics technology and financial leasing businesses. It is also an active investor in the robotics industry, having invested in companies such as Unitree Robotics, Galbot, Accelerate Evolution, Starship Technologies, StarDynasty, and Songyan Power. This background also means Kang Yu doesn't just focus on "who won" when observing the Games.

On one hand, robots are getting faster, more stable, and more autonomous year after year; on the other hand, a fall or a malfunction on the field often corresponds precisely to the problems that must be solved for robots to move toward mass production and real-world applications. "Looking at 2026 as it stands now, we believe this industry is just getting started," Kang Yu said. Although robots have made substantial progress this year in hardware motion control, product consistency, speed, battery life, and dexterous manipulation, there is still "a considerable distance" to go before they truly achieve "usefulness" and long-term reliable, stable operation. Robots are entering an era of reliability competition. One major change at this year's Games is the increased focus on autonomy. The event has placed greater emphasis on autonomous robots, with more and more teams prioritizing autonomous systems in both competitive and operational events. "Robots are gradually beginning to perceive and understand the world, and they will try to make decisions on their own," Kang Yu said. Compared to simply running 100 meters faster, this shift may be closer to the ultimate problem embodied intelligence needs to solve—whether robots can understand the real world and take autonomous action.

But even just running involves issues far more complex than a single result. Kang Yu noted that while audiences may see one robot break a record, some companies brought over a dozen robots to compete. This raises another question: if you scale from one robot to a dozen, how do you ensure consistency across each unit? This is precisely the challenge that must be addressed for future mass production. At the same time, whether damaged robots can be repaired, which components are most prone to failure, and why different hardware configurations produce different gaits are all issues that can be exposed intensively in the high-stress environment of the Games. "Any breakthrough for a robotics company is backed by countless training sessions," Kang Yu said. Failures on the field are equally meaningful. Some robots were "out of steam" after running 400 meters, others had problems before they even officially started, and some fell on the curves. The real world is also full of curves. Kang Yu explained that if robots are to enter daily life in the future, maintaining stability during continuous turning is critical; if components fall apart after just a few hundred meters, entering real-life scenarios could even pose safety risks. That's why she defines the Games as a "reliability test under the spotlight." The competition may expose robots' flaws in an unflattering way, but that is meaningful because these issues must ultimately be resolved before robots enter production and daily life. However, Kang Yu also repeatedly cautioned that the Games' rankings cannot simply be used to rank robotics companies. A competition by nature breaks down a robot's complex capabilities into individual metrics. The 100-meter dash tests speed, football tests coordination, and operational contests test dexterity—but a single capability does not equal a company's entire technological and industrial strength. Whether a robotics company is ready for mass production is difficult to demonstrate directly through the Games; it must be validated by customers and the market instead.

And "having the courage to compete" itself deserves recognition. She specifically mentioned a group of lesser-known participants. Near midnight one day, she passed the preparation venue for the football competition and found the lights still on. Some middle school students were still debugging their robots—some were writing summer homework while waiting for code to run; others were so tired they lay down on the grass to sleep, but as soon as they heard a teammate say "I think there's something wrong with this machine," they immediately sat up. "These people are the future of our robotics industry," she said. In her view, education and scientific research may not be the scenario where robotics most easily demonstrates commercial value, but they could be an underestimated link in the industry's development. These students who are learning to debug robots and modify algorithms today may become engineers or even entrepreneurs in the robotics industry in a few years. From the arena to the market, what does capital focus on? The Games also serve as a window for Shoucheng to observe the robotics industry chain. Kang Yu told reporters that Shoucheng arranges for investment managers to observe exactly "what went wrong" with robots on the field, and then analyze which links in the supply chain are worth investing in. Shoucheng's robotics investments are not limited to the main body. Previously, Shoucheng invested in Unitree Robotics, Tiangong, Accelerate Evolution, Songyan Power, and Galbot; in the model and embodied intelligence space, it has backed companies including Zibianliang, Galbot, Starship Technologies, and StarDynasty. In the upstream of the supply chain, she said Shoucheng is also seeking investment opportunities in areas such as joints, materials, data, computing power, storage, and interconnection. Taking joints as an example, Kang Yu said they not only account for a high proportion of a robot's value, but many of a robot's reliability issues also originate from the joints. As a result, Shoucheng has further invested in industry chain companies such as integrated joint manufacturers. She summarizes the investment approach as deploying around "physical AI." "Ultimately, everything must be connected from top to bottom before robots can achieve the embodied intelligence effects that we humans expect." Data is also an important direction. Kang Yu believes that embodied intelligence is currently in a state of "you can't make bricks without straw": real-world scenario data often still needs to be produced at a dedicated cost and has not yet reached the stage of large-scale natural data generation. This also relates to another debate in the industry: does embodied intelligence necessarily require a humanoid form? Do robots necessarily need bipedal legs and dexterous hands? Kang Yu's answer is that it is still far too early to bet on a single technology path. For example, Shoucheng invests in both bipedal robots and wheeled robots, as well as embodied intelligence companies of other forms. Embodiment is essentially the combination of AI and the physical world, so any form can be embodied—it doesn't have to be humanoid. But the humanoid form has its own logic. Since the vast majority of labor data in the real world is generated by humans, if robots have a form similar to humans, they may more easily leverage data naturally generated during human labor. She drew an analogy with autonomous driving: when cars drive on real roads, they can "incidentally" generate training data, rather than running dedicated data collection missions. If robots in the future can continuously generate data during real work, the data bottleneck for embodied intelligence could shift. "It's still too early to talk about convergence or betting on any specific technology line." Therefore, rather than betting on any specific robot form, Shoucheng currently emphasizes "betting on teams and betting on industry chain leaders," because the robotics industry will not always have good news. She noted that the robotics industry and its companies will inevitably go through cycles, the industry landscape is not yet determined, and consolidation in the future is certain. Some companies may be eliminated due to a lack of funding and an inability to generate their own cash flow, while industrial mergers and acquisitions, "big eating small," and upstream-downstream integration may also occur. One industry phenomenon is that some robotics companies that have received investment have begun investing in upstream companies such as dexterous hands, components, and data. Kang Yu does not consider this "going off track." For model companies, investing in dexterous hand or data collection companies may itself be a way to fill core capability gaps; for robotics companies gradually becoming "industry chain leaders," bringing core suppliers into their ecosystem through equity may become the next phase of competition. "As companies and the industry develop into the mid-term and beyond, it will definitely be an ecosystem-level strategy," Kang Yu predicted. Equity investment may be just the first step, and further capital operations such as acquisitions cannot be ruled out in the future. However, even with hot financing and accelerated IPOs, she still defines today's robotics industry as being in the "Spring and Autumn and Warring States period." Companies with true AI-native capabilities, killer applications, or "cliff-edge technological leadership" may not have emerged yet. And the breakthroughs on the field also reflect the current stage of the robotics industry. Robots are continuously pushing the boundaries of athletic capability, and companies are beginning to enter the capital markets, but there is still a long way to go before truly entering production and daily life scenarios, requiring them to overcome multiple challenges including reliability, data, cost, mass production, and commercialization.