Making coffee with head-mounted cameras: egocentric data capture for humanoid robot training
jonstephens85 · x · 2026-09-26
At Humanoids Summit, the author tried LightwheelAI's egocentric data capture system: head-mounted cameras plus wrist cameras recording human interaction from a first-person perspective.
- A simple task like making coffee becomes dozens of sub-actions for a robot: grab the cup, place it, pick up the pod, open the machine, insert the pod, close it, press the button
- Capturing and annotating these human demonstrations is a key way to create training data for robots learning longer-horizon real-world tasks
- Wearing all the gear while making coffee is admittedly awkward
More from Embodied
- Lighthouse extends Obsidian welding foundation model to run on any robotic platform — bennash · 2026-11-11
- Sipeed's MaixStation Dock PCB returns, 14-layer SOM still in fab — AIFlow_ML · 2026-11-11
- NAE EU-US Frontiers of Engineering symposium explores AI-driven materials, robotics, neuromorphic computing — ProfBuehlerMIT · 2026-11-11
- Paper link for FLEX: robots generalize language corrections into preferences — JafarNajafov · 2026-11-11
- MIT and NYU's FLEX lets robots generalize a single spoken correction into broad preferences — JafarNajafov · 2026-11-11
- AI lets robotics research thrive in labs with no robotics experts — 4310sy · 2026-11-11