← Leaderboard

ImageWAM

World-action
Shanghai Jiao Tong University · Jun 17, 2026

World-action model that repurposes image-editing models (FLUX.2, OmniGen2, Ovis-U1) to predict future frames and actions; from SJTU with Tencent Robotics X and Tsinghua co-authors.

Type
World-action
Parameters
—
Backbone
FLUX.2 [klein] 4B
Weights
Open
License
MIT
Commercial use
Allowed
Released
Jun 17, 2026
Robots
Franka Panda (sim), Aloha-AgileX (sim), Dobot X-Trainer, SO-101 (sim)
SO-101
–
LeRobot
–

Benchmark results

4 of 13 benchmarks
#13 of 2482.9%, self-reported by the authors, not a leaderboard entry
RoboTwin 2.0Near saturationSimulation
  • 93.4%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Jun 2026· self-reported· non-standard
93.4%, self-reported by the authors, not a leaderboard entrynot ranked
LIBEROSaturatedSimulation
#20 of 4698.4%, self-reported by the authors, not a leaderboard entry
SO-101 BenchLow-cost arm
#6 of 1143.0%, self-reported by the authors, not a leaderboard entry
Not reported on 9 benchmarks
Robustness drop: 98.4% on LIBERO → 82.9% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.

Adoption

updated Nov 17, 2026
GitHub stars
191
GitHub forks
12
Wrong or missing number? Report an error