LeHome Challenge Deep Dive: Sim-to-Real Robot Data Collection Practices
jonstephens85 · x · 2026-08-22
The blog post explores Part 3 of the LeHome Challenge, focusing on the Sim-to-Real gap in robotics. It details the author's home data-collection rig setup, domain randomization techniques for better generalization, and the human-in-the-loop DAgger interactive data collection method. These practices aim to improve robot operation and robustness in real-world home environments.
More from Embodied
- Richard Sutton Building 'Robot Kindergarten' in Beijing — RichardSSutton · 2026-08-22
- HKU Professor Zhang Fu's Startup Silicon Feather Bags Hundreds of Millions for Aerial Embodied AI — 量子位 · 2026-08-22
- Largest G1 Data Source Outside China: Lab Generates 200+ Hours of Teleoperation Data Weekly — DonJohnsonSays · 2026-08-22
- User shows Chinese supplier's custom humanoid robot parts — alexisgallagher · 2026-08-22
- Ex-Coohom exec launches consumer 3DGS camera 'Pebble' backed by Freedom Valley and SenseTime — 创业邦 · 2026-08-22
- NVIDIA Releases Newton 1.5: Upgrading Robot Physics Simulation — Olivier__OG · 2026-08-22