不跳舞的人形机器人，正尝试真正“干活”

Core Insights - The humanoid robot industry is focusing on practical applications beyond entertainment, with companies like Yushutech and Magic Atom announcing their participation in the Spring Festival Gala, while also facing criticism about the need for numerous dancing robots [1] - Companies are advancing humanoid robots into real-world scenarios, achieving significant milestones, such as Zivariable Robotics completing a 1 billion yuan financing and demonstrating autonomous food delivery using their self-developed VLA end-to-end model [1] - The end-to-end embodied large model technology aims to enable robots to make continuous decisions in real environments, but it is not a lightweight or highly certain technical path, requiring substantial funding and resources [2] Company Strategies - Zivariable Robotics is one of the early adopters of the end-to-end embodied large model technology, focusing on continuous decision-making from perception to action execution [1][2] - Sooteng Juchuang showcases a different approach with its VTLA-3D model, which integrates laser radar 3D point clouds and tactile feedback to enhance spatial understanding, reducing reliance on large-scale data during training [2] - The industry is divided into two main paths: one emphasizes immediate usability through multi-modal sensing, while the other focuses on long-term general intelligence through pure visual models [3] Engineering Approaches - A more engineering and delivery-oriented route is emerging, focusing on structured tasks and modular capabilities, allowing robots to perform reliably within defined task boundaries [3][4] - This engineering-focused approach offers high controllability and stability, making it suitable for semi-structured environments like industrial applications, but it struggles with adaptability to significant changes in tasks or environments [4] Industry Outlook - The humanoid robot industry is attempting to overcome several practical barriers to large-scale deployment, including safety coexistence, continuous operation, dexterous manipulation, and cost control [4] - Industry experts believe that while there is potential for breakthroughs in core capabilities within the next five years, achieving truly generalized humanoid robots will require time and accumulation of experience [5]