New paper benchmarks pi vs GR00T: how many demos does a VLA actually need to learn a task?
rahulshivchand · x · 2026-11-11
A new paper studies sample efficiency in vision-language-action models (VLAs). The authors ask two concrete questions: how many demonstrations does a VLA actually need to learn a new task, and how many training steps are required to reach a usable success rate?
They benchmark Physical Intelligence's pi model against NVIDIA's GR00T, comparing learning curves and final success rates under varying demonstration counts. The motivation: with enough demos VLAs reach high success rates, but how much is "enough" has not been systematically quantified.
More from Embodied
- Maestro paper: cheap Gemini 2.5 beats π₀.5 baselines at zero-shot robot control — dineshjayaraman · 2026-11-11
- OpenDDE: open-source all-atom drug discovery engine hits 51% on hard antibody-antigen benchmark — teortaxesTex · 2026-11-11
- Tesla demos Optimus walking inside a world model at CoRL, real policy in control — Scobleizer · 2026-11-11
- NUS Proposes WM-PPO: World-Model-Predicted Successor States Improve PPO Policy Learning — NationalUniversityofSingapore · 2026-11-11
- DexForesight Boosts Dexterous Manipulation Success From 49.4% to 60.5% With Implicit Spatial Foresight — Sharpa-Robotics · 2026-11-11
- NTU Study Finds Time-Indexed Memory Makes VLA Robots Mistake Pauses for Progress — NanyangTechnologicalUniversity · 2026-11-11