WLA-0 Makes Robots Think Before Acting
Gradio · x · 2026-07-12
Researchers introduced WLA-0, a unified World-Language-Action model that takes text, images, and robot states as inputs to predict subtasks before generating actions, integrating world modeling with language reasoning.
Key Results
- Achieved strong performance on long-horizon tasks: RoboTwin2.0 92.9%, RMBench 56.5%.
- A crucial ablation study showed that removing the "plan with text before acting" step drops the success rate from 56% to 17%, proving its importance.
- The model can learn directly from robot videos without requiring action annotations.
- Inference speed is roughly 40ms/step running on an RTX 5090.
Paper Highlights
- Aims to unify world modeling, language reasoning, and action synthesis within a single framework.
- Uses step-by-step prediction to enable the model to both understand the environment and generate executable actions.
- Suited for long-horizon robotic tasks requiring planning.
More from Embodied
- Johns Hopkins Launches Full-Stack Hands-on Robot Learning Class with SO-101 Arm Kits — _krishna_murthy · 2026-09-11
- SyncWorld: In-Context Robot World Model Simulates Unseen Views and Embodiments Zero-Shot — ChongZzZhang · 2026-09-11
- A 3D Pose Dataset for Dogs Released — ducha_aiki · 2026-09-11
- Swaayatt demos autonomous driving at 52 km/h on mountain roads, self-recovers after skid — sanjeevs_iitr · 2026-09-11
- AUAR's MicroFactory brings a deployable robotic wood-panel factory to the construction site — lukas_m_ziegler · 2026-09-11
- Musk: Cybercab certified at 165 Wh/mi, the most efficient production EV ever — elonmusk · 2026-09-11