文章 · 12 分钟
从视觉与指令到机器人动作：两条正在形成的路径
以训练目标、中间表征和推理时计算为线索，比较 VLM→VLA 与视频／世界模型→动作两条机器人控制路径。
Articles · 12 min
From Vision and Instructions to Robot Actions: Two Emerging Paths
A practical taxonomy for VLM-to-VLA and video/world-model-to-action systems, organized by training targets, internal representations, and inference-time computation.