During the 8th Hongqiao International Economic Forum's sub-forum themed "Collaboration for Innovative Progress in Humanoid Robotics," Wang Xingxing, the visionary founder of Unitree Robotics, highlighted that the principal obstacles confronting agent robots currently revolve around models and data. Regarding the model aspect, current architectures fall short in terms of generalization abilities, posing challenges in adapting to a wide array of situations. On the data front, the criteria for data collection and quality assessment are unclear, coupled with a shortage of top-notch data. He underscored the necessity for models and data to undergo simultaneous optimization, rather than merely aiming for an increase in scale. Wang Xingxing projected that the "ChatGPT moment" for embodied artificial intelligence could materialize within the next 1-2 years, facilitating the accomplishment of 80% of tasks in unfamiliar settings through voice or textual instructions. Additionally, he disclosed that Unitree Robotics is intensifying its data acquisition efforts via full-body dynamic teleoperation technology and is delving into world models grounded in video generation. Nevertheless, obstacles persist, including elevated computational requirements and complexities in aligning different data modalities.