The second World Humanoid Robot Games in Beijing offered a useful split-screen view of humanoid robotics: rapid progress in tightly defined athletic tasks, and conspicuous fragility when those systems left their narrow operating envelope. According to Ars Technica, viral clips from the August 22–26 event showed sprinting humanoids beating the human 100-meter record held by Usain Bolt. Other videos showed the other side of the demonstration: robots unable to stop cleanly, crashing into barriers, falling, breaking at the waist in showers of sparks, and in a few cases catching fire. The Beijing event reprised the spectacle of the inaugural 2025 games with contests built for attention: robot running, kung fu or dance routines, kickboxing, soccer, and table tennis. But Ars reports that this year’s games also included more practical tasks, including washing and hanging laundry, where the robots appeared to remain well behind ordinary human working speeds. That contrast matters because the competition is not just entertainment. Ars says the organizers included Chinese state media and the Beijing city government, and frames the event as a showcase for the commercial potential of humanoid robots as China’s robotics industry develops and tests such systems. Dipam Patel, a Purdue University computer science PhD candidate and research assistant at the US Army DevCom Army Research Lab, told Ars that moving from trailing human record-holders last year to surpassing human records in the 100-meter sprint and standing high jump this year indicates meaningful progress in whole-body control. In those events, the robot has to coordinate legs, arms, torso, and joints toward a specific objective. But Patel also cautioned that the performances do not show general-purpose autonomy. Ars reports that many robots were engineered for individual tasks, with sprinting machines optimized to run straight through the finish rather than behave like a rounded autonomous system that can slow down, recover, and continue safely. In that context, a robot breaking apart after the line is not incidental; it reveals what the system was, and was not, designed to do. The most commercially relevant signal may be the least viral one. Humanoid robots that can sprint, jump, or kickbox are impressive engineering demos, but Ars’s account suggests household-task performance remains the harder benchmark for buyers who care about useful labor. For investors and operators, the gap between record clips and laundry tasks is the story: narrow physical control is improving, while broadly useful autonomy is still not established by this event. Who benefits: Robotics companies that can stand out in visible competitions benefit from publicity and, as Patel told Ars, potential investor attention. Event organizers also get a platform to showcase China’s robotics sector. Who's exposed: Companies implying that sprinting or jumping demos translate directly into general-purpose humanoid capability are exposed by the failures and slower chore performance Ars describes. Buyers should treat single-task records as engineering milestones, not proof of broad autonomy.