Robotics Needs a New 'Capture' Technology: Why Kinematics Data Is the Missing Scaling Input
andrew_n_carr · x · 2026-10-13
Andrew Carr lays out a structural argument about robotics' data bottleneck:
- Every profitable modality went through a capture-generate cycle: the internet captured language, IDEs captured code, cameras captured video—data largely existed via capture before generation.
- Robotics requires kinematics data, which today only comes from expensive motion-capture rigs or deployed hardware—neither scales to the levels needed to hit performance thresholds.
- Since scaling laws likely dominate even more in robotics, this implies the field needs a new kind of capture technology that translates an existing modality into the shape robotics needs.
More from Embodied
- Two gaps in sequence: Koltun's is representation, Malik's is information — contact isn't on the screen — le_james94 · 2026-10-13
- Shuran Song: robot data is not a quantity problem — a week of pick-and-place logs 600GB yet is nearly useless elsewhere — le_james94 · 2026-10-13
- Shuran Song's UMI recipe: handheld gripper + GoPro, then add sound, force and hands one channel at a time — le_james94 · 2026-10-13
- LeCun's world model pitch: predict representations, not pixels, and plan via MPC-style cost minimization — le_james94 · 2026-10-13
- Malik's three levels of manipulation: LLMs live on top, robots stall on forces and torques — le_james94 · 2026-10-13
- OpenAI 'dots' hardware spotted in the wild, stuck at UK customs — SamuelAlbanie · 2026-10-13