Landed in Austin for #CoRL2026! Can't wait to see old friends and meet new ones.
Find me Thu (11/12) morning at two workshop posters:
โ GNR (done at Meta Reality Labs) @ Scaling H2R
๐ EGR (done at MERL) @ Beneath the Policy
DMs open, always up for coffee! Details ๐
2
21
419
๐ Breaking Modality Entanglement in Vision-Language-Action Policies
Work from MERL | Everything Beneath the Policy workshop
yy-gx.github.io/EGR/
1
17
โ Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation
Work from Meta Reality Labs | Scaling H2R workshop
yy-gx.github.io/GNR/
3
56
๐ฆพ Scaling dexterous robot demonstrations from human data!
๐ Excited to share Generative Neural Retargeting (GNR) from Meta Reality Labs!
๐ฅ 223k robot demos across 3.3k objects, with dense contact force labels!
๐ yy-gx.github.io/GNR/
๐ arxiv.org/abs/2610.12440
๐ Why Generative?
GNR learns a distribution of physically feasible robot trajectories, enabling test-time sampling to explore multiple solutions and improve retargeting success.
Huge thanks to our amazing team at Meta Reality Labs Research! ๐ @DechenGao @ewhitmire
1
127
๐ค GNR scales data along two axes:
๐ Diversity: Large-scale, contact-rich demonstrations generated with high-fidelity physics simulation using SuperDex (projectsuperdex.com/).
๐ฏ Difficulty: High-precision, long-horizon manipulation, including NIST assembly and ShapeFilter.
2
221
