← Leaderboard
LDA-1B
World-actionGalbot et al. · Feb 12, 2026
Latent-dynamics action model trained on mixed-quality data. It self-reports gains of up to 21% on contact-rich and 48% on dexterous tasks and was a RoboDojo launch baseline.
- Type
- World-action
- Parameters
- 1B
- Backbone
- Frozen Qwen3-VL-4B + latent dynamics action model
- Weights
- Open
- License
- Apache-2.0
- Commercial use
- Allowed
- Released
- Feb 12, 2026
- Robots
- ARX X5 (sim)
- SO-101
- –
- LeRobot
- –
- Inputs
- –
- VRAM
- –
Benchmark results
1 of 13 benchmarksRoboDojo (sim)Simulation
- Official leaderboard ↗Jul 6, 2026
#47 of 540.5%
Not reported on 12 benchmarks
Wrong or missing number? Report an error