Ablations show human data lifts robot success from 15% to 62.5%
AIFlow_ML · x · 2026-10-26
AIFlowML describes a capture rig any lab can copy today: a GoPro on the neck at robot-camera height, PICO trackers for 24 body joints, and hand pose estimated from images — 29s per human demo vs 100s per robot demo. The key ablation: full recipe hits 62.5% success; removing the HumanVerse mid-training stage drops it to 22.5%; removing all human-data training leaves 15.0%. Same robot, same team, same post-training — human data is the difference.
Related event: Alibaba's λ0 Trains Humanoid Robot with 500 Hours of Human Video(4 posts)→
More from Embodied
- On-device VLM goes production: GenieX runs Liquid AI's LFM-VL live at PyTorch Conference — PyTorch · 2026-10-26
- LeCun co-authored Temporal Straightening: straightening latent paths boosts gradient-based planning success rates — ylecun · 2026-10-26
- Humanoid robots' DRAM demand may 20x, but it's still just 0.04% of 2026 supply — zephyr_z9 · 2026-10-26
- Microduck team publishes deep dives: relocalizes within 2cm using a 64-pixel depth sensor — mishig25 · 2026-10-26
- Robot's elbow-assisted get-up move goes viral for its humanlike fluidity — iamfakhrealam · 2026-10-26
- Chinese eVTOL EHang EH216-S completes nearly 100,000 flights across 23 countries — LoganGrasby · 2026-10-26