Ablation: freezing the VLM hurts π models far more than GR00T
rahulshivchand · x · 2026-11-11
A repeated post within the same thread: freezing the VLM backbone during fine-tuning costs π models 23-56 success points vs at most 6 for GR00T, restating that GR00T N1.7's stronger zero-shot doesn't translate to faster adaptation. Largely duplicates earlier thread posts.
More from Embodied
- Beni robot, #1 personal robot on Kickstarter, officially launches with self-recovery and 4K60 — Scobleizer · 2026-11-11
- Berkeley's Social Humanoid teaches Unitree G1 socially responsive interaction — berkeley_ai · 2026-11-11
- Robotics is not solved: frontier models are ~6 orders of magnitude off real assembly lines — ZeeshanZiaML · 2026-11-11
- MacBooks do have a hinge angle sensor—verified with Codex, prompt included — lipeng0820 · 2026-11-11
- DynaRobotics robot demo at CoRL sparks 'I want one at home' reactions — JasonMa2020 · 2026-11-11
- Humanoid Factory Deploy Stalls: Teleop Intervention Every 6 Minutes After 15 Weeks — ZeeshanZiaML · 2026-11-11