本文由 AI 分析生成
建立時間： 2026-05-30 來源： https://x.com/Klajd_Lika/status/2051425222192980413
Summary
Klajd Lika explains why locomotion and manipulation are fundamentally different control problems using a physics thought experiment. A falling body (locomotion analogy) from 1 meter gives ~450 control cycles at 1 kHz. A 1mm falling object (manipulation analogy) gives only ~4–5 cycles — a ~100x difference in available reaction time. This means cameras, processors, and software designed for locomotion are inappropriate for agile manipulation.
Klajd Lika 以自由落體實驗解釋移動控制與操作控制的根本差異：從1公尺落下約有450個控制週期，從1毫米落下只有4–5個。這種100倍的差距使得針對移動設計的感測器和電腦對精細操作幾乎無效。文章也指出視覺感測對毫米尺度事件的天生局限性。
Key Points
- Locomotion: robot IS the object, goal is preventing falls
- Manipulation: robot controls an external object with sub-mm precision
- The 100x time-pressure gap requires 100x faster sensors, processors, bandwidth
- Cameras can’t capture the force/micro-deflection dynamics during contact
- Vision captures state before and after contact — the critical contact interval is invisible
Insights
The framing is unusually crisp: it’s not that manipulation is “harder,” it’s that it operates in a different physics regime that demands different sensing modalities (tactile, proprioceptive) rather than better cameras. This explains why the locomotion boom (Boston Dynamics, etc.) didn’t automatically transfer to dexterous manipulation. The post also honestly flags its own physics approximation.
Connections
Raw Excerpt
“This isn’t just an engineering inconvenience — it means the cameras, computers, and software developed for locomotion are, in a meaningful sense, the wrong tools for agile in-hand manipulation.”