At the ATEC2025 Real-World Extreme Challenge, held at The Chinese University of Hong Kong (CUHK), competing robots faced a grueling test: mud, stairs, swaying bridges, and unstructured outdoor environments. Unlike typical robotics contests staged indoors, this competition deliberately placed intelligent robots in the unpredictable conditions of the real world—a move organizers say is essential to bridge the gap between laboratory demonstrations and actual deployment.
Professor Yunhui Liu, Co-Chair of ATEC2025 and Fellow of the Hong Kong Academy of Engineering, framed the event as a necessary stress test. As he noted, “This year’s competition sought to answer a fundamental question: Can robots truly leave the lab and adapt to our unpredictable world?”
The vision behind ATEC2025 reflects a broader shift in artificial intelligence—one echoed by founding member Ant Group. Ant Group’s support for ATEC2025 is rooted in a core belief that the future of artificial general intelligence lies in the deep integration of machine intelligence with the physical world.
A Test of Three Core Capabilities
Out of 392 teams from universities and industries worldwide that participated in the online qualifiers, only 13 advanced to the two-day final: the Real-World Extreme Challenge. They were tasked with completing four real-world challenges: waste sorting, autonomous plant watering, field orienteering, and bridge crossing.
These tasks were designed to evaluate what Prof. Liu calls the “Three Core Capabilities of Robotics”: locomotion, manipulation, and environmental modification.
“We emphasize combining movement and operation,” Liu explained in media interviews. “Previously, many robot competitions focused mainly on walking—even marathons. But here, robots must also grasp objects, pour water, and change their environment to complete tasks.”
Progress Through Failure
The difficulty was evident. Many teams struggled, with even basic grasping requiring numerous attempts. “Some of the best teams needed 10 or 20 tries to succeed,” Liu noted.
However, such incomplete performances were expected in this inaugural real-world format. “This is the first time we’ve done something like this. Failure is normal,” Liu said. “What matters is that we see progress—some teams did fully complete the tasks after repeated trials.”
Liu emphasized that ATEC2025 is not about polished demos but real-world robustness. “Videos of robots you see online are often edited highlights after dozens of failed attempts,” he said. “Here, we see the raw process. Walking on flat ground may look reliable, but on uneven terrain or narrow stairs, failure rates rise significantly.”
Looking to the Future
While the winning team will receive a $150,000 prize, Liu made clear that ATEC’s primary aim is not commercialization but foundational advancement. “We hope this inspires young people to join the field and pushes the community to solve hard problems together,” he said. Though partnerships with investors could emerge later, “this competition is fundamentally about testing robotic intelligence.”
Looking ahead, Liu believes meaningful real-world deployment remains years away. “Robots today handle simple services like food delivery, but tasks like cooking or laundry in a home setting require far greater reliability, safety, and dexterity,” he cautioned. “It won’t happen in two or three years—maybe not even in five.”
“The goal,” Liu said, “is for robots to reliably perform even 20–30% of basic human tasks, like sorting trash or watering plants, with 90% success. That alone would be transformative.”
Still, he sees ATEC2025 as a critical step. By anchoring evaluation in human-relevant capabilities—measuring performance not just in accuracy but in functional equivalence to human actions—the competition aims to explore new benchmarks for the field.
With strong support from academia, industry, and the public sector, and leveraging the hardware and talent ecosystem of the Greater Bay Area, ATEC may well become a recurring proving ground where the future of embodied AI is forged—not in simulation, but in soil, wind, and rain.
