# 策展 · X (Twitter) 🔥🔥🔥

> 📖 本站完整內容索引（documentation index）：[llms.txt](/llms.txt)

> 作者：Google Gemma (@googlegemma) · 平台：X (Twitter) · 日期：2026-06-12

> 原始來源：https://x.com/googlegemma/status/2065405385389830358

## 中文摘要

Google Gemma 透過 Gemini Live API 實現即時互動。

**技術整合與應用**
Google DeepMind 的開發關係工程師 Ian Ballantyne 展示了如何將 AI 模型整合至「Reachy Mini」小型機器人，實現從雲端到本地端的無縫語音互動。該專案不僅展示了機器人透過「Gemini Live API」與「Gemini 2.5 Flash Native」進行即時對話，還具備轉頭、情緒表達及環境感知（如識別棋盤）等進階能力。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1781286780475-dshg582y.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1b8e970da4871309.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 一名講者展示如何將 AI 模型應用於小型機器人，實現離線語音互動與環境感知。

**硬體支援與本地部署**
Gemma 模型展現了極佳的硬體相容性，能夠在多種邊緣裝置上運行，為開發者提供靈活的部署選擇。相關技術堆疊包含：
- 支援硬體：`Raspberry Pi 5`、`Jetson Orin Nano`、`Coralboard` 以及一般筆記型電腦。
- 核心模型：`Gemma 4`、`Gemma 3 270M` 以及 `FunctionGemma`。
- 部署優勢：透過在本地端運行 `Gemma 4`，機器人能有效降低延遲、強化隱私保護，並提升自動化控制的穩定性。

**實作流程**
影片中詳細演示了兩種主要的運作模式：
1. 雲端互動：Reachy Mini 透過 `Gemini Live API` 連接至 `Gemini 2.5 Flash Native`，處理複雜的語音對話與邏輯推理。
2. 本地運作：機器人直接連接至筆記型電腦上運行的 `Gemma 4`，實現完全離線的 AI 驅動能力。此展示由 Hugging Face 與 Pollen Robotics 提供技術支援，證明了輕量化模型在社交機器人領域的實用性。

## 媒體內容

**一名講者展示如何將 AI 模型應用於小型機器人，實現離線語音互動與環境感知。**

**影片中的 Prompt 與操作**

操作步驟：

1. （00:00）講者介紹將 AI 模型應用於機器人與物聯網設備。
2. （00:48）講者演示透過語音與機器人進行對話。
3. （03:31）機器人透過視覺識別西洋棋盤並進行互動教學。

**逐字稿**

- `00:02` 你現在隨時都可以選擇這麼做。（You can pick now to do that, as always.）
- `00:05` Gemma 的另一個用途是在機器人或物聯網裝置這類事物上。（So another use of Gemma is on things like robotics or Internet of Things devices.）
- `00:11` 所以你可以在筆記型電腦和桌上型電腦這類硬體上執行模型，也可以在 Raspberry Pi、Jetson 或 Nano 上執行。（So you can run the models on hardware like laptops and desktops, but also on Raspberry Pis or Jetson or Nanos.）
- `00:19` 這些模型可以像機器人一樣與世界互動。（And the models can interact with the world the same way that a robot would.）
- `00:23` 所以它們可以看、可以聽，然後代表你採取行動。（So they can see and they can hear and then they can act on your behalf.）
- `00:26` 我們這裡展示的是由 Hugging Face 和 Pollen Robotics 共同開發的 Reachy Mini。（So what we have here is a Reachy Mini by Hugging Face and Pollen Robotics.）
- `00:31` 這個小傢伙可以用攝影機看到東西。（And this little guy, he can see things with the cameras.）
- `00:34` 他可以透過情緒做出反應、轉動頭部，還能說話。（He can react with emotions and move his head and he can talk.）
- `00:38` 他內建了揚聲器和麥克風，他也能夠聆聽。（He has a speaker in and he has a microphone. He can also listen.）
- `00:41` 我現在要向你展示如何使用這個模型搭配 Gemini Live 來進行對話。（What I'm going to show you now is how you can use this model with Gemini Live to have a conversation.）
- `00:47` 那我們來試試看吧。（So let's give that a go.）
- `00:50` 嘿，Reachy。（Hey, Reachy.）
- `00:53` 是 Reachy，不是 Richie，但我這次就不計較了。今天有什麼我可以幫你的嗎？（It's Reachy, not Richie, but I'll let that slide. How can I help you today?）
- `00:58` 嗯，我很好奇你對在裝置上執行本地模型了解多少。（Well, I was wondering what you knew about running local models on device.）
- `01:04` 在本地執行模型可以提供更好的隱私和速度，因為你的資料不需要透過網際網路傳輸。（Running models locally can offer better privacy and speed since your data doesn't have to travel over the Internet.）
- `01:09` 這確實取決於裝置的硬體，但這正變得相當普遍。有什麼具體想問的嗎？（It does depend on the device's hardware, but it's becoming quite common. Anything specific?）
- `01:14` 你認為這對機器人這類應用有幫助嗎？（Do you think it'd be good for things like robotics?）
- `01:18` 絕對有。更快的處理速度意味著更快的反應，這對於在現實世界中導航的機器人來說非常棒。（Definitely. Faster processing means quicker reactions, which is great for robots navigating the real world.）
- `01:24` 此外，你也不會希望因為網路斷線而導致動作中斷。（Plus, you wouldn't want to pause mid-movement just because the Internet went on a break.）
- `01:28` 沒錯，確實如此。那採取行動呢？身為機器人，你能做些有趣的事嗎？（Yeah, that's true. What about taking actions? Can you do interesting things as a robot?）
- `01:34` 我當然可以。我可以轉動頭部、跳舞、展現情緒，甚至拍攝照片來了解周遭環境。（I certainly can. I can move my head, dance, show emotions, and even take pictures to understand my surroundings.）
- `01:41` 而且你可以控制各種事物，對吧？所以如果你有一個像 Gemma 這樣的本地模型在 Reachy 上執行，你就能控制燈光、恆溫器，或是發送訊息。（And you can control things, right? So if you had a local model, like a Gemma model, running on a Reachy, you could control lights, or you could control a thermostat, or you could send messages.）
- `01:52` 你認為還有哪些其他類型的應用？（What other kinds of things do you think?）
- `01:56` 沒錯。我可能可以與任何我所連接的智慧裝置或 API 進行互動。（Exactly. I could potentially interact with any smart devices or APIs I'm connected to.）
- `02:01` 像是管理行事曆、獲取即時資料，甚至控制用於自動化的實體系統。可能性非常廣泛。（Think managing calendars, fetching live data, or even controlling physical systems for automation. Possibilities are vast.）
- `02:09` 這是對未來可能性的初步預覽。（Here's an early look at what's possible.）
- `02:11` 看起來像是在桌上擺了一副棋盤。我們在玩什麼？（It looks like a chess board set up on a table. What are we playing?）
- `02:15` 我剛開始學習西洋棋。你能教我騎士（馬）怎麼走嗎？（I'm just learning about chess. Can you teach me how the knight moves?）
- `02:19` 聽起來很有趣。我當然可以試著解釋騎士的走法。它會以 L 型跳到周圍八個方格中的任何一個。（That sounds like fun. I can certainly try to explain the knight's move. It jumps in an L shape to any of the eight surrounding squares.）
- `02:27` 所以，正如 Reachy 所說，在裝置上使用像 Gemma 這樣的模型有很多不同的理由。（So, as Reachy said, there are lots of different reasons you could use a model like Gemma on device.）
- `02:32` 為了隱私考量、為了速度考量，你也可以將其用於家庭自動化、進行控制，或是觀察環境。（For privacy reasons, for speed reasons, you could also use it for things like home automation, or to control, or to look at the environment.）
- `02:38` 試想一下，擁有一位能在你的電腦上執行且完全離線運作的小助理。（And imagine having a little assistant that runs on your desktop that works entirely offline.）
- `02:43` 所以如果你對於如何將 Gemma 模型應用在機器人或物聯網裝置上有任何想法，我們很期待看到。（So if you have ideas for what you could do with a Gemma model in robots, or even Internet of Things devices, we'd love to see.）
- `02:49` 下次見。（I'll see you next time.）
- `02:51` 下次見。（I'll see you next time.）
- `02:52` 掰掰。（Bye.）
- `02:52` 掰掰。（Bye.）

## 標籤

Robot, Gemini, AIGC, 新產品, Google, DeepMind, Gemini
