← Leaderboard

X-VLA

VLA
Tsinghua AIR et al. · Oct 11, 2025

Cross-embodiment VLA that uses soft prompts per embodiment. It reports results on LIBERO, SimplerEnv and CALVIN, and the RoboTwin team lists it as a baseline.

Type
VLA
Parameters
900M
Backbone
Florence-2-large
Weights
Open
License
Apache-2.0
Commercial use
Allowed
Released
Oct 11, 2025
Robots
WidowX (Bridge), Google Robot, Franka Panda (sim), Aloha-AgileX (sim), AgileX, AIRBOT
SO-101
–
LeRobot
–

Benchmark results

6 of 13 benchmarks
#30 of 406.5%, self-reported by the authors, not a leaderboard entry
RoboTwin 2.0Near saturationSimulation
#18 of 2344.5%
CALVIN ABC→DNear saturationSimulation
#8 of 114.43 / 5, self-reported by the authors, not a leaderboard entry
LIBEROSaturatedSimulation
#23 of 4698.1%, self-reported by the authors, not a leaderboard entry
SimplerEnv WidowXSaturatedSimulation
#3 of 1995.8%, self-reported by the authors, not a leaderboard entry
#9 of 123.3%, self-reported by the authors, not a leaderboard entry
Not reported on 7 benchmarks

Adoption

updated Nov 17, 2026
GitHub stars
763
GitHub forks
79

Trained on

Runs on

Wrong or missing number? Report an error