The whistle blew, the scoreboard froze at 9:0. Accelerated Evolution’s humanoid robot T1 lifted the championship trophy at the RoboCup German Open adult division final. Cheng Hao, former captain of Tsinghua University’s Vulcan Team, fulfilled a decade-long ambition that began during his student days.
RoboCup, known as the “Robot World Cup,” saw Tsinghua Vulcan Team—founded by robotics professor Zhao Mingguo—claim victory with Cheng’s T1 robots. The achievement marked a full-circle moment for Cheng, who as a student had led the team to a Teen-Size division third-place finish in earlier competitions.
From Software to Hardware: A Founder’s Journey
After graduating, Cheng’s career initially diverged from robotics—he founded productivity app “Morning & Night Calendar,” later acquired by ByteDance where he became Vice President of Feishu (Lark). The 2023 ChatGPT 3.0 breakthrough, however, sharpened his vision for combining large language models with robotic motion control. “This historical juncture offers unparalleled potential for humanoid robots,” Cheng observed. “They’ll evolve into multimodal systems that perceive and influence the physical world—requiring new hardware platforms.”
This insight led to Accelerated Evolution’s founding in August 2023, assembling former Vulcan Team members for R&D and recruiting Professor Zhao as Chief Scientist. Within a year, their T1 prototype was competing—and winning—on the global stage.
T1’s RoboCup Dominance
At March’s RoboCup German Open final, T1 robots demonstrated startling sophistication—one “predicted” an opponent’s defensive route, maneuvering around a taller German robot to score. The 9:0 shutout victory forced rule changes mid-tournament as most competitors’ robots couldn’t withstand physical collisions.
This followed a challenging July 2024 debut where overheating joints and communication failures required 10 backup robots for a single match. “We embraced ‘compensate height with skill’ as our motto,” Cheng said. Post-match analysis drove material science breakthroughs—testing new “skin” composites by dropping 20kg balls from two meters—and algorithm refinements that reduced stand-up time from 10 seconds to one.
Beijing’s Tech Ambassador
By August 2024, T1 became Beijing’s “city tech icon,” showcased at the Zhongguancun Forum where dozens performed synchronized push-ups, hosted panels, and played football—demonstrating collective AI intelligence. Mass production began in October, with the 100th unit delivered this March.
Technically, T1 employs reinforcement learning where thousands of virtual robots train in simulated environments. “It’s like a child learning to walk—but compressed into hours equivalent to a decade of human practice,” Cheng explained. Reward-punishment mechanisms guide training before deploying models to physical units, coordinating full-body joints for complex actions like powerful kicks.
The Future of Embodied AI
“Football is an efficient testing ground requiring vision, decision-making, and motion control—skills transferable to real-world tasks,” Cheng noted. While acknowledging current limitations in embodied AI models, he believes they represent the industry’s future direction.
Backed by Zhongguancun Science City’s Pre-A funding, Cheng envisions humanoid robots reaching billions of units across factories, offices, and homes—a trajectory he compares to computers evolving from luxury items to ubiquity. “Like in Klara and the Sun, they’ll eventually wait in storefronts for adoption,” he mused, referencing Kazuo Ishiguro’s novel about artificial companions.
