跨系列概念图
同一方法会出现在不同系列里。图上是对照入口，不是互斥分类。点节点会跳到代表篇；文章页底部还有更细的相关篇。
flowchart TD
PG[策略梯度 / PPO] --> RLHF[RLHF / InstructGPT]
RLHF --> DPO[DPO]
RLHF --> GRPO[GRPO / R1]
Q[Q-learning / DQN] --> SAC[SAC / TD3]
WM[世界模型 / Dreamer] --> VLA[VLA / OpenVLA]
IM[模仿 / GAIL / DAgger] --> DMP[DMP]
ATT[注意力 / Transformer] --> RAG[RAG]
ATT --> CLIP[CLIP / LLaVA]
CLIP --> VLA
PLAN[A* / RRT] --> CTRL[控制与轨迹]
EST[Kalman / SLAM] --> SYS[完整系统]
DMP --> CTRL
click PG "rl/27-ppo.html"
click RLHF "llm/054-instructgpt.html"
click DPO "llm/075-dpo.html"
click GRPO "rl/52-grpo.html"
click Q "rl/04-q-learning.html"
click SAC "rl/30-sac.html"
click WM "rl/40-dreamer.html"
click VLA "robot-learn/082-openvla.html"
click IM "robot-learn/006-dagger.html"
click DMP "robot-learn/001-dmp.html"
click ATT "llm/015-transformer.html"
click RAG "llm/034-rag.html"
click CLIP "llm/043-clip.html"
click PLAN "robot-classic/003-astar.html"
click EST "robot-classic/002-kalman.html"