Ant Group Co., Ltd., which has yet to directly venture into robot manufacturing, is placing early bets on promising players in the field.
On December 6, at Lingnan Sports Ground of The Chinese University of Hong Kong, there were no flawlessly choreographed robot performances under spotlights. Instead, young teams gathered by grassy fields, stone steps, and suspension bridges, observing their creations. A quadruped robot hesitated as it navigated a swaying bridge, while a robotic arm repeatedly missed grabbing a plastic bottle on the grass. Though seemingly clumsy, these scenes excited Professor Liu Yunhui, Chair of the Expert Committee and Fellow of the Hong Kong Academy of Engineering Sciences—because no remote controls or pre-programmed scripts were used. Every action relied on the robots’ autonomous perception, decision-making, and execution in unpredictable outdoor environments.
This was the finals of the fifth ATEC Tech Elite Challenge. As a key organizer, Ant Group aimed to pull robots out of lab-controlled settings and into the harsh real world. For the first time, the competition was held entirely outdoors on natural terrain, with rules explicitly encouraging zero remote operation—pushing robots to evolve from remote-controlled "tools" into autonomous agents.
Achieving "zero remote operation" demands robots independently complete full-chain tasks—perception, analysis, decision-making, and execution—amid real-world uncertainties. Any misstep could derail the mission, requiring extreme robustness in sensing, decision-making intelligence, and system stability.
"The competition tackles a core question: Can robots truly leave labs and adapt to our complex world?" said Liu. "We want to push them from ‘demo-viable’ to ‘application-reliable’ through extreme challenges."
Thus, the contest was deliberately set in unstructured wilderness: shifting light, uneven terrain, wind-blown leaves. Robots had to complete four tasks—orienteering, bridge-crossing, autonomous plant-watering, and waste-sorting—each riddled with complexities.
"Algorithms perfected in labs face endless surprises in the wild," admitted Zhu Chengrui, captain of Zhejiang University’s Wongtsai team. During plant-watering, robots had to identify pots and flowers while navigating paths and adjusting grip strength—any error meant failure. Zhu noted that integrating robots into daily life remains daunting: "Even a misplaced pillow forces them to re-learn."
Liu emphasized that the competition tested three core robotic capabilities—locomotion, manipulation, and environmental adaptation—with real-world execution posing exponentially higher difficulties. By placing challenges on hills, grass, steps, and bridges, Ant Group sought to expose robots’ true weaknesses.
"If problems aren’t real, they won’t drive real progress," stated Ant’s Head of Technology Strategy. "Only ‘real problems’ show the industry what to tackle next."
Behind this "real-world extreme challenge" lies Ant’s broader ambition in embodied AI: through a competition targeting technical pain points, it aims to discover and invest in top robotics innovators like Unitree and Agibot.
Ant’s focus stems from embodied AI’s impending breakout, currently bottlenecked by three hurdles: environmental perception/cognition, intelligent decision-making/response, and hardware/computing power. The contest pushes young participants to break barriers, with real-world testing offering clearer direction.
If Unitree founder Wang Xingxing pioneered quadruped robotics commercialization with XDog, Ant now seeks the next pioneers who can shift embodied AI from "demo" to "Day One" usability.
In the digital realm, Ant’s Lingguang and AQ already lead in agent technology. But the ultimate form of AGI/ASI lies in merging machine intelligence with the physical world—transitioning from "data cognition" to "environment interaction" and "action execution."
Ant recognizes this leap requires an ecosystem, not solo efforts. Over 70% of contestants hailed from top-tier universities, including overseas institutions, embodying the same innovative drive Wang Xingxing had when building his first robot dog in a dorm.
Zhejiang University’s Wongtsai team won for exceptional autonomous control. But for Ant, the real goal isn’t victory—it’s finding the next "Wang Xingxing" to shatter technical barriers. Liu predicts household "nanny-level" robots may take 5+ years. In this marathon of "body" and "mind," embodied AI’s future may hinge on rising after each fall.
Comments