Robot ablation: removing human-data training drops success from 62.5% to 15%
AIFlow_ML · x · 2026-10-26
A stark ablation for robot training, holding the robot, team, and post-training constant: the full recipe hits 62.5% success.
- Removing the HumanVerse mid-training stage: success collapses to 22.5%
- Removing all human-data training entirely: only 15.0%
The takeaway: human-data training — and where it sits in mid-training — is decisive, accounting for most of the performance. Project page, preprint, and full code are linked for verification.
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