← Leaderboard

Qwen-RobotManip

VLA
Alibaba Qwen · Jun 16, 2026

Qwen Team VLA that pairs Qwen3.5-4B with a flow-matching action expert and is evaluated mainly on out-of-distribution settings. The team says it has no plan to release the weights.

Type
VLA
Parameters
—
Backbone
Qwen3.5-4B
Weights
Closed
License
–
Commercial use
Unclear
Released
Jun 16, 2026
Robots
Franka Panda (sim), Aloha-AgileX (sim), AgileX ALOHA, Franka, UR, ARX
SO-101
–
LeRobot
–

Benchmark results

4 of 13 benchmarks
#18 of 2635.9%, self-reported by the authors, not a leaderboard entry
#5 of 2489.7%, self-reported by the authors, not a leaderboard entry
RoboTwin 2.0Near saturationSimulation
  • 67.9%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Jun 2026· self-reported, clean-only training· non-standard
67.9%, self-reported by the authors, not a leaderboard entrynot ranked
LIBEROSaturatedSimulation
#8 of 4699.1%, self-reported by the authors, not a leaderboard entry
Not reported on 9 benchmarks
Robustness drop: 99.1% on LIBERO → 89.7% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.

Adoption

updated Nov 17, 2026
GitHub stars
189
GitHub forks
9
Wrong or missing number? Report an error

Built on Qwen-RobotManip