Shanghai-based AgiBot has cracked a major industrial automation challenge: teaching robots complex manufacturing tasks in just 10 minutes. The startup's hybrid approach combines human teleoperation with reinforcement learning to rapidly train two-armed robots for production lines, potentially transforming how factories operate across China and beyond.
AgiBot just deployed something that could reshape manufacturing: robots that learn new factory tasks in 10 minutes flat. The Shanghai startup's breakthrough combines human guidance with AI reinforcement learning, creating machines that adapt to production line changes almost as quickly as their human counterparts. The company's G2 humanoid robot is already working at Longcheer Technology's production line, handling components for smartphones and VR headsets with a speed that has industry observers taking notice. "The robot deployed at the Longcheer plant takes components from a machine that performs testing, then places them onto a production line," explains AgiBot representative Yuheng Feng. It's the kind of precise, repetitive work that typically requires weeks of traditional robot programming. But AgiBot's approach flips that script entirely. The secret lies in what they call Real-World Reinforcement Learning - a system that starts with human workers guiding robots through tasks via teleoperation, then lets AI take over to refine and perfect the movements. Chief scientist Jianlan Luo, who previously conducted cutting-edge robotics research at UC Berkeley, brought this human-in-the-loop methodology to industrial scale. The implications extend far beyond a single production line. China's manufacturing dominance creates perfect conditions for this technology to flourish. The country already operates more industrial robots than every other nation combined, according to the International Federation of Robotics. The Chinese government's latest five-year plan explicitly calls for AI and robotics-driven economic growth, setting the stage for massive investment. "Rapid learning is important because production lines often change from one week to the next, or even during the same production run," Feng notes. Traditional industrial robots require extensive reprogramming for each task variation. AgiBot's machines adapt in real-time, maintaining pace with human workers as requirements shift. The training process itself requires significant human investment. AgiBot runs a dedicated robotic learning center where workers teleoperate robots to generate training data. This mirrors a growing trend - to perform manual tasks that serve as robot training data. Jeff Schneider, a roboticist at Carnegie Mellon University specializing in reinforcement learning, confirms that "AgiBot is using cutting-edge techniques, and should be able to automate tasks with high reliability." He suggests other robotics companies are likely experimenting with similar approaches for manufacturing applications. The competitive landscape is heating up globally. In the US, - a cofounded by researchers who worked on the same UC Berkeley project as Luo - is developing similar robot learning algorithms. , a Carnegie Mellon spinout, recently demonstrated that work across different physical robot forms. But China's massive manufacturing infrastructure provides unique advantages. The country offers rapid prototyping capabilities, massive production scale, an eager market for robot labor, and abundant workers to help train AI models. One unnamed US robotics entrepreneur told Wired that while American competitors don't keep him awake at night, "Chinese robotics firms" definitely do. AgiBot's success could signal a broader shift in global manufacturing competitiveness. As production becomes increasingly automated and AI-driven, countries with the infrastructure to deploy and refine these systems rapidly will gain significant advantages. The startup's ability to train robots in minutes rather than weeks could fundamentally change how factories respond to market demands.
