自变量完成10亿元融资 字节、红杉出手

Group 1 - The core viewpoint of the news is that Zibianli has successfully completed a 1 billion RMB A++ round of financing, led by top investment institutions including ByteDance and Sequoia China, marking a significant milestone for the company and the AI investment landscape in China [1] - Zibianli, founded in December 2023, focuses on developing embodied intelligent general models to achieve general robotics, positioning itself uniquely in the market as the only embodied intelligence company backed by major internet giants like Meituan and Alibaba [1][2] - The company’s founder and CEO, Wang Qian, emphasizes that the next phase of competition in embodied intelligence will revolve around the foundational models and their evolutionary capabilities, driven by the complexity of the physical world [1] Group 2 - Zibianli's self-developed WALL-A model integrates VLA and world models, significantly enhancing the robot's zero-shot generalization ability in unstructured environments, which is crucial for executing mobile tasks [2] - The company has heavily invested in data since its inception, establishing a closed-loop iteration of hardware, data, and models, and has developed various data collection devices to validate data and achieve breakthroughs in model performance [2] - Demonstrations show that Zibianli's robots can autonomously complete tasks like food delivery and cardboard recycling, even in challenging conditions, by leveraging their foundational model's generalization capabilities and causal reasoning [4] Group 3 - Zibianli has also developed its own hardware, launching high-performance robot bodies named "Quantum One" and "Quantum Two," which include self-researched core components such as robotic arms and controllers [4] - Wang Qian believes that the short-term development of embodied intelligence relies on algorithmic advantages, while mid-term success will depend on data advantages, and long-term success will hinge on product advantages [4] - The company is actively exploring service closure practices in various open service scenarios, aiming to deepen its accumulation in three key areas: model algorithms, data, and product development [4]