About 12,000 human athletes participated in a half marathon in Beijing, but the spotlight was on 21 humanoid robots that also took part in the event. This was touted as the first instance where humans and bipedal robots raced concurrently, albeit on separate tracks. Only six of the robots managed to complete the course, substantially trailing behind the human runners.
The fastest robot, Tiangong Ultra, developed by UBTech, completed the race in two hours and 40 minutes. This was after its operators changed batteries three times during the run, and it fell once along the way. In contrast, the human runners had a minimum completion time of three hours and ten minutes, making Tiangong Ultra barely eligible for a human participation award.
Most of the other robots faltered soon after the race began, vanishing from the live broadcast. Alan Fern, a robotics professor from Oregon State University, pointed out that the primary focus of researchers has been on improving the functionality of robots in various environments rather than speed running. He noted that while hardware for these robots had become increasingly reliable, the AI technology driving them had seen little advancement since 2021.
Fern’s predictions came true during the race, as nearly all robots struggled with falls and overheating issues, necessitating frequent replacements. The event, while drawing significant national pride and interest—such as human runners stopping for selfies with Tiangong Ultra—also underscored the limitations of China’s humanoid robot sector.
Several Chinese companies showcased impressive humanoid robots, gaining international attention earlier in the year. For instance, Unitree, which didn’t officially participate, saw its robots running the half marathon while being directed by different institutions.
Despite their visually appealing designs and various entertaining abilities like dancing, the performance of these humanoid robots during the half marathon highlighted a gap in practical utility. Instead of competing with human runners, Fern emphasized that the relevant benchmarks should focus on handling diverse real-world tasks autonomously.
The humanoid robots varied in design, with the smallest measuring just 2 feet and 5 inches tall, while the Tiangong Ultra stood at five feet nine. Each robot adhered to bipedal requirements, allowing their creators to experiment with strategies for the race. Many wore makeshift footwear, while others donned protective gear like knee pads to mitigate damage from falls.
Tiangong Ultra and another robot, N2 from Noetix Robotics, managed to maintain a slow but steady pace. In contrast, a robot named Huanhuan moved at a snail’s pace for only a few minutes and displayed uncontrollable head movements. Another contraption, Shennong, which looked like a fusion of different concepts, spun around unexpectedly, leading to chaotic scenes.
Duct tape emerged as an unconventional but effective tool for addressing issues during the race, such as fixing dislodged heads and making temporary shoes. Human operators, often multiple per robot, acted as guides and caretakers, dealing with operational challenges while keeping pace themselves.
As the race unfolded, spectators observed the exhaustion of the human operators, who faced the dual tasks of managing their robots and completing the run alongside them. While some robots showcased various tricks, including dances and backflips, the event itself amplified the visible limitations of humanoid robotics.
Notably, the Noetix robot Xuanfeng Xiaozi experienced multiple breakdowns yet managed to complete the race, demonstrating the resilience despite the evident design flaws. As it approached the finish line, it wore cooling pads and wobbling steps—a moment that drew cheers as it crossed, emphasizing both the struggles and small triumphs of these humanoid robots.
