← Leaderboard

Rho

VLA
Microsoft Research · Sep 29, 2026

Open-weight 5B VLA family for data-light bimanual adaptation, released as Rho-base plus checkpoints for three dual-arm platforms, LIBERO and RoboEval. It adds an online-adaptation module that learns from as few as 15 corrected episodes by steering the frozen action expert’s noise inputs.

Type
VLA
Parameters
5.2B
Backbone
Phi-Phy (4.68B Phi-family VLM with SigLIP 2) + flow-matching action expert
Weights
Open
License
MIT
Commercial use
Allowed
Released
Sep 29, 2026
Robots
YAM Box, UR AI Trainer, Franka FR3 Duo
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
LIBEROSaturatedSimulation
  • 97.9%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Sep 2026· n = 6000· non-standard
97.9%, self-reported by the authors, not a leaderboard entrynot ranked
Not reported on 12 benchmarks
Wrong or missing number? Report an error