← Leaderboard

NORA

VLA
SUTD · Apr 28, 2025

3B VLA from SUTD’s DeCLaRe Lab, fine-tuned from Qwen2.5-VL-3B on 970k Open X-Embodiment demonstrations with the FAST+ action tokenizer. A NORA-Long variant predicts 5-action chunks. The LIBERO numbers are self-reported after per-suite fine-tuning.

Type
VLA
Parameters
3B
Backbone
Qwen2.5-VL-3B
Weights
Open
License
MIT
Commercial use
Allowed
Released
Apr 28, 2025
Robots
WidowX 250, Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
RGB · L
VRAM
8.3GB

Benchmark results

2 of 13 benchmarks
#27 of 2840.5%, self-reported by the authors, not a leaderboard entry
LIBEROSaturatedSimulation
#58 of 7087.9%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Robustness drop: 87.9% on LIBERO → 40.5% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.
Wrong or missing number? Report an error

Built on NORA