循环Transformer跑在微控制器上实时生成游戏
推荐指数 68.0 NO. 017 · 2026.10.30
发布2026/10/29Score53Comments5
简介
研究者用两个小型神经网络在树莓派Pico 2 W上同时生成和玩游戏：世界模型预测下一帧，策略网络选择动作。关键创新是用单层Transformer循环复用权重，而非堆叠多层，将内存占用压到MCU能承载的范围。
编辑判断
这个项目真正有趣的不是跑在MCU上，而是验证了循环Transformer的压缩极限：用同一组权重迭代多次，能否逼近深层网络的表达能力。之前DeepMind的Looped Transformer论文主要在学术benchmark上验证，这是首次在实时交互场景落地。
对端侧AI团队来说，这提示了一个设计空间：如果你的模型受限于SRAM而非Flash，循环架构可能比知识蒸馏更划算。作者提到Pico 2 W的FPU是关键——没有硬件浮点加速，循环展开的延迟会压垮实时性。做MCU部署的人需要重新评估自己的硬件选型是否卡在了这个隐性门槛上。
社区反馈
正面 5 条评论
核心争论：循环Transformer的MCU部署与世界模型能力哪个更值得关注
This is incredibly cool! Fascinating to see looped transformers be necessary for mini-Doom, but not for simpler games.
Honestly fascinating. I almost feel like you'd be better off avoiding any discussion of the fact that this runs on a Raspberry Pi Pico 2 W because it takes away from the crazy world model capabilities (of course, the fact that this does run on a Pico 2 makes the whole thing much more impressive). Th
Very cool. I didn't know NNs were capable of such determinism, that's very interesting. Is this a similar effect to overfitting? Also, it's the same model for all the games, right?