Survey

graph TB subgraph VLM["VLM as Driving Explainer"] V1[🖼️ Vision
Input] --> VLM1[🤖 VLMs
Processing] VLM1 --> E1[💬 Explain
Q&A
Description] style V1 fill:#e3f2fd style VLM1 fill:#fff3e0 style E1 fill:#fff8e1 end

subgraph Modular["Modular VLA for AD"]
    V2[🖼️ Multimodal<br/>Vision] --> VLM2[🤖 VLMs<br/>Processing]
    VLM2 --> IR[🔄 Intermediate<br/>Representation]
    IR --> AH[⚙️ Action Head]
    AH --> TC1[🎯 Trajectory<br/>Control]
    style V2 fill:#e3f2fd
    style VLM2 fill:#fff3e0
    style IR fill:#f3e5f5
    style AH fill:#e8f5e8
    style TC1 fill:#fce4ec
end

subgraph EndToEnd["End-to-end VLA for AD"]
    V3[🖼️ Multimodal<br/>Vision] --> VLM3[🤖 VLMs<br/>Processing]
    VLM3 --> A1[🚗 Action<br/>Output]
    style V3 fill:#e3f2fd
    style VLM3 fill:#fff3e0
    style A1 fill:#e0f2f1
end

subgraph Augmented["Augmented VLA for AD"]
    V4[🖼️ Multimodal<br/>Vision] --> RT[🧠 Reasoning VLMs<br/>& Tool-use Agents]
    RT --> A2[🚗 Action<br/>Output]
    style V4 fill:#e3f2fd
    style RT fill:#f3e5f5
    style A2 fill:#e0f2f1
end

%% 演进箭头
VLM -.-> Modular
Modular -.-> EndToEnd
EndToEnd -.-> Augmented

%% 样式定义
classDef titleStyle fill:#f9f9f9,stroke:#333,stroke-width:2px,color:#333
class VLM,Modular,EndToEnd,Augmented titleStyle