虎嗅APP·2025-06-28 14:23Core Viewpoint - The article discusses the advancements in robotics powered by Google's Gemini AI technology, highlighting its ability to perform tasks without continuous internet connectivity and its potential impact on the robotics industry [2][6][24]. Group 1: Gemini Robotics On-Device Model - The Gemini Robotics On-Device model was released on June 24, enabling robots to operate offline, which is beneficial for applications sensitive to latency and ensures robustness in intermittent or zero connectivity environments [6][7]. - This model aims to enhance robots' adaptability to new tasks and environments, addressing issues such as dexterous manipulation, fine-tuning for new tasks, and low-latency inference based on local operation [9][20]. - In performance comparisons, Gemini Robotics On-Device showed significant improvements over previous offline models, although slightly lower than the flagship Gemini Robotics model [14][16]. Group 2: Task Performance and Adaptability - The model demonstrated strong visual, semantic, and behavioral generalization capabilities, successfully completing tasks like placing blocks and opening drawers based solely on natural language commands [13][20]. - After being trained with 50 to 100 examples, the model exhibited impressive adaptability, allowing developers to fine-tune it for new tasks quickly [20]. - The model was tested on dual-arm robots, successfully executing complex tasks that require precision and dexterity, such as folding clothes and industrial assembly [20][22]. Group 3: Industry Implications and Comparisons - The introduction of Google's offline AI robots has the potential to change the game in the robotics sector, although there are questions about how it compares to Tesla's robot designs and Meta's work [24]. - The article emphasizes the diversity and richness of technological approaches in the robotics and embodied intelligence field, all aiming to enable AI to establish genuine causal understanding in the physical world [24].