ModelsMulti-source

Google DeepMind 展示 Gemini 4 Hydrogen:照片生成 3D 仿真场景,约一小时搭建机械臂分拣系统

Why we picked it

DeepMind 官方演示了一条很实用的机器人开发路径:拍照丢进 MuJoCo 就能生成 3D 仿真,聊天里改动作,直接部署到真机械臂,官方说一小时做出了个传送带分拣机。做机器人或产线自动化的朋友可以照着这条流程试试,不用先买硬件。

Google DeepMind 在 X 上发布演示视频,介绍 Gemini 4 Hydrogen 在机器人开发中的新用法:上传一张照片或一段提示词,即可在物理仿真引擎 MuJoCo 中生成对应的 3D 场景。用户随后通过对话方式调整场景中的物体和运动逻辑,模型将改动直接部署到真实机械臂上,官方称从白板草图到可运行的传送带分拣系统耗时约一小时。该模型已可通过 Antigravity、Google AI Studio 和 Gemini API 三种渠道试用。这套流程把仿真验证前置到动硬件之前,开发者可以在虚拟环境里先试错再上真机,降低机器人原型开发的试错成本。对比传统机器人开发中手写仿真环境和调参的流程,这种方式把场景搭建和逻辑调整都收敛到一次对话交互里,对机械臂、传送带分拣这类工业自动化的快速原型验证有直接参考价值。

Image · Google DeepMind
Body

对机器人开发者和小型自动化团队来说,这条演示给出了一条低成本的工作流路径:不需要先购置硬件,也不需要熟悉 MuJoCo 的底层建模接口,用照片就能起步。企业买家如果在做产线分拣、抓取类的方案评估,可以先用 Gemini API 复现这条流程再做硬件投入决策。

Original · Google DeepMind

You can even test ideas before touching hardware. Upload a photo or prompt to build a 3D scene in @MuJoCo , adjust objects and movements in chat, and deploy straight to your arm, going from a whiteboard sketch to a working conveyor sorter in about an hour. Try Gemini 4 Hydrogen in @Antigravity , @GoogleAIStudio , or the Gemini API → goo.gle/4eHJVez

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Gemini 4 Hydrogen, MuJoCo, Google AI Studio, robot simulation.

Source attribution: curated from Google DeepMind (@GoogleDeepMind). Cross-source corroboration: 4 independent sources have covered this development (including Patrick Loeber, theverge, Artificial Analysis), confirming industry-wide signal strength.