硬氪专访|商汤联创王晓刚带队具身智能新业务,要帮机器人重新理解真实世界
3 6 Ke·2025-12-15 01:32

Core Insights - SenseTime has transitioned from visual AI to generative AI, with its revenue from generative AI expected to reach 2.4 billion yuan in 2024, increasing its share from 34.8% in 2023 to 63.7% [1] - The establishment of Daxiao Robotics, led by co-founder Wang Xiaogang, aims to address the practical challenges of AI in the physical world through a human-centric approach [2][4] - The AI industry is evolving from "digital intelligence" to "physical intelligence," necessitating a significant transformation for established AI companies [3] Company Developments - SenseTime reported a net loss of 1.162 billion yuan in the first half of 2025, a 50% decrease year-on-year, while continuing to increase R&D investment [4] - Daxiao Robotics aims to create integrated soft and hard products that address real-world needs, moving beyond merely developing models [10][12] - The company is focusing on the robotics sector as a critical breakthrough point for expanding its business and achieving vertical integration [5] Technological Innovations - Daxiao Robotics proposes a human-centric paradigm that emphasizes understanding human interactions with the physical world, utilizing diverse data sources [7][24] - The new "Kairos 3.0" world model integrates multi-modal understanding and behavior prediction, allowing for a comprehensive understanding of physical interactions [19][21] - The model's architecture includes multi-modal understanding, synthesis networks, and predictive capabilities, enabling robots to execute complex tasks in dynamic environments [22][23] Market Trends - The Chinese embodied intelligence market is projected to exceed 800 billion yuan in 2024, with hundreds of startups entering the field [10] - Daxiao Robotics aims to leverage its understanding of user pain points and needs to accelerate the industrialization of embodied intelligence [5][24] - The company anticipates that the logistics sector will be a key area for scaling its solutions, while household applications will require longer development cycles due to safety concerns [28]