Google DeepMind 发布实时、照片级世界模型 Genie 4,可用自然语言操控,新增原生音频并改进碰撞、空气动力学等物理表现。文章认为其对机器人的价值在于提供无限可变、可操控的训练环境,目前正与 Gemini 4 Argon 和 Gemini Robotics 配合,仿真 Apptronik 的 Apollo 人形机器人下跳棋,将任务拆解为实时步骤。
AI 生成摘要 · 以原文为准
Robots are about to get an infinite world to practice in.
Google DeepMind unveiled Genie 4, a real-time, photorealistic world model you steer with natural language, now with native audio and far better physics (collisions, aerodynamics).
For robotics, the interesting part isn't the pretty worlds. It's this: infinitely variable, steerable training environments for robots and agents.
They're already pairing Genie 4 with Gemini 4 Argon and Gemini Robotics to simulate Apptronik's Apollo humanoid playing checkers, breaking the task into real-time steps.
Sim-to-real, but the sim generates itself.
来源:The Humanoid Hub · x.com