Google 推出 Gemini 4 Argon,欲以单一模型覆盖机器人全流程
原标题:Google wants Gemini 4 Argon to be the only AI model robots need
AI 摘要
Google 发布面向机器人任务的 Gemini 4 Argon 模型,宣称用单一模型覆盖机器人控制、生成交互式 3D 训练环境以及评估训练数据。演示中它可从照片或草图构建仿真环境,让机器人先在虚拟世界练习再部署到真实硬件,例如让 Boston Dynamics Spot 练习投掷易拉罐,并生成超 4500 条训练片段使真实任务成功率从 56% 提升至 75%。Google 称其在 ERQA 具身推理基准上以 87% 领先其他前沿模型,但在 Artificial Analysis 的 AA-Robotics 真实机械臂控制测试中仅列第三(55 分),落后 GPT-6 Astra 和 GPT-6.1 Sol。Argon 目前通过 Google 付费开发者计划提供,演示仍依赖详细的人类提示与引导。
正文节选
Google wants Gemini 4 Argon to be the only AI model robots need Key Points - Gemini 4 Argon is Google's attempt at a single AI model that controls robots, generates interactive 3D training environments, and evaluates training data. - From simple inputs like photos or sketches, the system builds simulations where robots can practice movements before real-world deployment. - The demos still require detailed human prompting and guidance. Argon is available through Google's paid developer access pro