理想汽車於 2026 年 9 月 22 日發布 ME Brain 1.0 系統、ME VLM 視覺語言模型,以及連結任務理解與動作生成的 ME U0。[3][5]
ME U0 在 LIBERO 與 LIBERO Plus 的平均成功率分別為 99.1% 和 81.8%,但這些成績經過各基準測試提供的監督訊號適配,不能當成陌生現實任務的零樣本成功率。[3]
M100 晶片上的 188 毫秒數字,測量的是 4B 版 ME VLM 的推論預填階段,不是機器人從看見變化到完成動作的總反應時間。[1]
What did Li Auto’s Foundation Model team release in its September 22, 2026, MachEmbodied embodied-AI trilogy, and how do ME-Brain-1.0, ME-VLAI-generated illustration; not a photograph of Li Auto’s hardware or demonstrations.
AI 提示詞
Create a landscape editorial hero image for this Studio Global article: What did Li Auto’s Foundation Model team release in its September 22, 2026, MachEmbodied embodied-AI trilogy, and how do ME-Brain-1.0, ME-VL. Article summary: Li Auto’s Foundation Model team presented three complementary pieces on September 22: ME-Brain-1.0 as an embodied-agent system, ME-VLM as its vision-language cognitive core, and ME-U0 as a model for understanding scenes . Topic tags: general, academic, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, char
openai.com
理想汽車的 Foundation Model 團隊在 2026 年 9 月 22 日一口氣發布三項具身 AI 成果:ME-Brain-1.0、MachEmbodied-VLM(ME-VLM)與 MachEmbodied-U0(ME-U0)。如果把機器人完成任務想成「記住經驗、看懂眼前情況,再決定並執行動作」,這三項成果分別對應其中不同環節。不過,將它們並列發布,並不等於已證明整套系統能在同一台邊緣裝置上自主、穩定地運作。35