← Leaderboard

TinyVLA

VLA
Midea Group & ECNU · Sep 19, 2024

Compact VLA family that pairs small multimodal models with a diffusion action head for fast inference without large-scale robot pretraining. Only the VLM backbones are released; it was a RoboDojo launch baseline.

Type
VLA
Parameters
1.3B
Backbone
Llava-Pythia (0.4B / 0.7B / 1.3B) + diffusion policy head
Weights
Unknown
License
MIT
Commercial use
Unclear
Released
Sep 19, 2024
Robots
ARX X5 (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
#52 of 540.1%
Not reported on 12 benchmarks
Wrong or missing number? Report an error